2009-08-01 17:27:05 +01:00
|
|
|
/**************************************************************************
|
|
|
|
*
|
2010-08-14 18:02:47 +01:00
|
|
|
* Copyright 2009-2010 VMware, Inc.
|
2009-08-01 17:27:05 +01:00
|
|
|
* All Rights Reserved.
|
|
|
|
*
|
|
|
|
* Permission is hereby granted, free of charge, to any person obtaining a
|
|
|
|
* copy of this software and associated documentation files (the
|
|
|
|
* "Software"), to deal in the Software without restriction, including
|
|
|
|
* without limitation the rights to use, copy, modify, merge, publish,
|
|
|
|
* distribute, sub license, and/or sell copies of the Software, and to
|
|
|
|
* permit persons to whom the Software is furnished to do so, subject to
|
|
|
|
* the following conditions:
|
|
|
|
*
|
|
|
|
* The above copyright notice and this permission notice (including the
|
|
|
|
* next paragraph) shall be included in all copies or substantial portions
|
|
|
|
* of the Software.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
|
|
|
|
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
|
|
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
|
|
|
|
* IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
|
|
|
|
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
|
|
|
|
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
|
|
|
|
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
|
|
|
*
|
|
|
|
**************************************************************************/
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @file
|
|
|
|
* Helper
|
|
|
|
*
|
|
|
|
* LLVM IR doesn't support all basic arithmetic operations we care about (most
|
|
|
|
* notably min/max and saturated operations), and it is often necessary to
|
|
|
|
* resort machine-specific intrinsics directly. The functions here hide all
|
|
|
|
* these implementation details from the other modules.
|
|
|
|
*
|
|
|
|
* We also do simple expressions simplification here. Reasons are:
|
|
|
|
* - it is very easy given we have all necessary information readily available
|
|
|
|
* - LLVM optimization passes fail to simplify several vector expressions
|
|
|
|
* - We often know value constraints which the optimization passes have no way
|
|
|
|
* of knowing, such as when source arguments are known to be in [0, 1] range.
|
|
|
|
*
|
|
|
|
* @author Jose Fonseca <jfonseca@vmware.com>
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
2012-12-04 13:38:52 +00:00
|
|
|
#include <float.h>
|
|
|
|
|
2019-08-28 00:06:21 +01:00
|
|
|
#include <llvm/Config/llvm-config.h>
|
|
|
|
|
2009-08-12 10:35:50 +01:00
|
|
|
#include "util/u_memory.h"
|
2009-08-03 22:31:08 +01:00
|
|
|
#include "util/u_debug.h"
|
2009-10-25 11:48:17 +00:00
|
|
|
#include "util/u_math.h"
|
2009-09-29 13:58:58 +01:00
|
|
|
#include "util/u_cpu_detect.h"
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2009-08-03 22:24:01 +01:00
|
|
|
#include "lp_bld_type.h"
|
2009-08-12 02:34:55 +01:00
|
|
|
#include "lp_bld_const.h"
|
2010-11-30 23:07:52 +00:00
|
|
|
#include "lp_bld_init.h"
|
2009-08-07 01:16:59 +01:00
|
|
|
#include "lp_bld_intr.h"
|
2009-08-31 10:00:38 +01:00
|
|
|
#include "lp_bld_logic.h"
|
2009-10-22 18:28:57 +01:00
|
|
|
#include "lp_bld_pack.h"
|
2010-09-09 12:09:44 +01:00
|
|
|
#include "lp_bld_debug.h"
|
2012-12-04 13:38:52 +00:00
|
|
|
#include "lp_bld_bitarit.h"
|
2009-08-01 17:27:05 +01:00
|
|
|
#include "lp_bld_arit.h"
|
2013-07-09 04:45:55 +01:00
|
|
|
#include "lp_bld_flow.h"
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2013-12-06 06:28:25 +00:00
|
|
|
#if defined(PIPE_ARCH_SSE)
|
|
|
|
#include <xmmintrin.h>
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifndef _MM_DENORMALS_ZERO_MASK
|
|
|
|
#define _MM_DENORMALS_ZERO_MASK 0x0040
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifndef _MM_FLUSH_ZERO_MASK
|
|
|
|
#define _MM_FLUSH_ZERO_MASK 0x8000
|
|
|
|
#endif
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2011-07-20 22:53:08 +01:00
|
|
|
#define EXP_POLY_DEGREE 5
|
2010-08-14 18:02:47 +01:00
|
|
|
|
2012-04-05 20:34:11 +01:00
|
|
|
#define LOG_POLY_DEGREE 4
|
2010-08-14 18:02:47 +01:00
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate min(a, b)
|
|
|
|
* No checks for special case values of a or b = 1 or 0 are done.
|
2013-07-16 18:06:24 +01:00
|
|
|
* NaN's are handled according to the behavior specified by the
|
|
|
|
* nan_behavior argument.
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2009-08-02 17:54:53 +01:00
|
|
|
static LLVMValueRef
|
|
|
|
lp_build_min_simple(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
2013-07-16 18:06:24 +01:00
|
|
|
LLVMValueRef b,
|
|
|
|
enum gallivm_nan_behavior nan_behavior)
|
2009-08-01 17:27:05 +01:00
|
|
|
{
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-02 17:54:53 +01:00
|
|
|
const char *intrinsic = NULL;
|
2012-07-17 21:41:29 +01:00
|
|
|
unsigned intr_size = 0;
|
2009-08-02 17:54:53 +01:00
|
|
|
LLVMValueRef cond;
|
|
|
|
|
2010-08-09 22:15:08 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
assert(lp_check_value(type, b));
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
/* TODO: optimize the constant case */
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if (type.floating && util_get_cpu_caps()->has_sse) {
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
if (type.width == 32) {
|
|
|
|
if (type.length == 1) {
|
|
|
|
intrinsic = "llvm.x86.sse.min.ss";
|
|
|
|
intr_size = 128;
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (type.length <= 4 || !util_get_cpu_caps()->has_avx) {
|
2009-08-02 17:54:53 +01:00
|
|
|
intrinsic = "llvm.x86.sse.min.ps";
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
intr_size = 128;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
intrinsic = "llvm.x86.avx.min.ps.256";
|
|
|
|
intr_size = 256;
|
|
|
|
}
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
if (type.width == 64 && util_get_cpu_caps()->has_sse2) {
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
if (type.length == 1) {
|
|
|
|
intrinsic = "llvm.x86.sse2.min.sd";
|
|
|
|
intr_size = 128;
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (type.length == 2 || !util_get_cpu_caps()->has_avx) {
|
2009-08-02 17:54:53 +01:00
|
|
|
intrinsic = "llvm.x86.sse2.min.pd";
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
intr_size = 128;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
intrinsic = "llvm.x86.avx.min.pd.256";
|
|
|
|
intr_size = 256;
|
|
|
|
}
|
2009-08-08 22:57:03 +01:00
|
|
|
}
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (type.floating && util_get_cpu_caps()->has_altivec) {
|
2021-04-29 21:55:30 +01:00
|
|
|
if (nan_behavior == GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN) {
|
2013-11-11 14:29:25 +00:00
|
|
|
debug_printf("%s: altivec doesn't support nan return nan behavior\n",
|
|
|
|
__FUNCTION__);
|
|
|
|
}
|
2012-11-22 17:03:11 +00:00
|
|
|
if (type.width == 32 && type.length == 4) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vminfp";
|
|
|
|
intr_size = 128;
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
} else if (util_get_cpu_caps()->has_altivec) {
|
2013-07-16 18:06:24 +01:00
|
|
|
intr_size = 128;
|
|
|
|
if (type.width == 8) {
|
|
|
|
if (!type.sign) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vminub";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vminsb";
|
|
|
|
}
|
|
|
|
} else if (type.width == 16) {
|
|
|
|
if (!type.sign) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vminuh";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vminsh";
|
|
|
|
}
|
|
|
|
} else if (type.width == 32) {
|
|
|
|
if (!type.sign) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vminuw";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vminsw";
|
|
|
|
}
|
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (intrinsic) {
|
2013-07-16 18:06:24 +01:00
|
|
|
/* We need to handle nan's for floating point numbers. If one of the
|
|
|
|
* inputs is nan the other should be returned (required by both D3D10+
|
|
|
|
* and OpenCL).
|
|
|
|
* The sse intrinsics return the second operator in case of nan by
|
|
|
|
* default so we need to special code to handle those.
|
|
|
|
*/
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse && type.floating &&
|
2021-04-29 21:55:30 +01:00
|
|
|
nan_behavior == GALLIVM_NAN_RETURN_OTHER) {
|
2014-08-30 03:58:58 +01:00
|
|
|
LLVMValueRef isnan, min;
|
|
|
|
min = lp_build_intrinsic_binary_anylength(bld->gallivm, intrinsic,
|
2013-07-16 18:06:24 +01:00
|
|
|
type,
|
|
|
|
intr_size, a, b);
|
2021-04-29 21:55:30 +01:00
|
|
|
isnan = lp_build_isnan(bld, b);
|
|
|
|
return lp_build_select(bld, isnan, a, min);
|
2013-07-16 18:06:24 +01:00
|
|
|
} else {
|
|
|
|
return lp_build_intrinsic_binary_anylength(bld->gallivm, intrinsic,
|
|
|
|
type,
|
|
|
|
intr_size, a, b);
|
|
|
|
}
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2013-07-16 18:06:24 +01:00
|
|
|
if (type.floating) {
|
|
|
|
switch (nan_behavior) {
|
|
|
|
case GALLIVM_NAN_RETURN_OTHER: {
|
|
|
|
LLVMValueRef isnan = lp_build_isnan(bld, a);
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_LESS, a, b);
|
|
|
|
cond = LLVMBuildXor(bld->gallivm->builder, cond, isnan, "");
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
}
|
|
|
|
break;
|
2013-11-11 14:29:25 +00:00
|
|
|
case GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN:
|
2013-07-16 18:06:24 +01:00
|
|
|
cond = lp_build_cmp_ordered(bld, PIPE_FUNC_LESS, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
2014-08-29 04:12:00 +01:00
|
|
|
case GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN:
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_LESS, b, a);
|
|
|
|
return lp_build_select(bld, cond, b, a);
|
2013-07-16 18:06:24 +01:00
|
|
|
case GALLIVM_NAN_BEHAVIOR_UNDEFINED:
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_LESS, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_LESS, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_LESS, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
}
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2016-04-04 00:05:33 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_fmuladd(LLVMBuilderRef builder,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b,
|
|
|
|
LLVMValueRef c)
|
|
|
|
{
|
|
|
|
LLVMTypeRef type = LLVMTypeOf(a);
|
|
|
|
assert(type == LLVMTypeOf(b));
|
|
|
|
assert(type == LLVMTypeOf(c));
|
2019-09-06 08:29:23 +01:00
|
|
|
|
2016-04-04 00:05:33 +01:00
|
|
|
char intrinsic[32];
|
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.fmuladd", type);
|
|
|
|
LLVMValueRef args[] = { a, b, c };
|
|
|
|
return lp_build_intrinsic(builder, intrinsic, type, args, 3, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate max(a, b)
|
|
|
|
* No checks for special case values of a or b = 1 or 0 are done.
|
2013-07-16 18:06:24 +01:00
|
|
|
* NaN's are handled according to the behavior specified by the
|
|
|
|
* nan_behavior argument.
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2009-08-02 17:54:53 +01:00
|
|
|
static LLVMValueRef
|
|
|
|
lp_build_max_simple(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
2013-07-16 18:06:24 +01:00
|
|
|
LLVMValueRef b,
|
|
|
|
enum gallivm_nan_behavior nan_behavior)
|
2009-08-01 17:27:05 +01:00
|
|
|
{
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-02 17:54:53 +01:00
|
|
|
const char *intrinsic = NULL;
|
2012-07-17 21:41:29 +01:00
|
|
|
unsigned intr_size = 0;
|
2009-08-02 17:54:53 +01:00
|
|
|
LLVMValueRef cond;
|
|
|
|
|
2010-08-09 22:15:08 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
assert(lp_check_value(type, b));
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
/* TODO: optimize the constant case */
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if (type.floating && util_get_cpu_caps()->has_sse) {
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
if (type.width == 32) {
|
|
|
|
if (type.length == 1) {
|
|
|
|
intrinsic = "llvm.x86.sse.max.ss";
|
|
|
|
intr_size = 128;
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (type.length <= 4 || !util_get_cpu_caps()->has_avx) {
|
2009-08-02 17:54:53 +01:00
|
|
|
intrinsic = "llvm.x86.sse.max.ps";
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
intr_size = 128;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
intrinsic = "llvm.x86.avx.max.ps.256";
|
|
|
|
intr_size = 256;
|
|
|
|
}
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
if (type.width == 64 && util_get_cpu_caps()->has_sse2) {
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
if (type.length == 1) {
|
|
|
|
intrinsic = "llvm.x86.sse2.max.sd";
|
|
|
|
intr_size = 128;
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (type.length == 2 || !util_get_cpu_caps()->has_avx) {
|
2009-08-02 17:54:53 +01:00
|
|
|
intrinsic = "llvm.x86.sse2.max.pd";
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
intr_size = 128;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
intrinsic = "llvm.x86.avx.max.pd.256";
|
|
|
|
intr_size = 256;
|
|
|
|
}
|
2009-08-08 22:57:03 +01:00
|
|
|
}
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (type.floating && util_get_cpu_caps()->has_altivec) {
|
2021-04-29 21:55:30 +01:00
|
|
|
if (nan_behavior == GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN) {
|
2013-11-11 14:29:25 +00:00
|
|
|
debug_printf("%s: altivec doesn't support nan return nan behavior\n",
|
|
|
|
__FUNCTION__);
|
|
|
|
}
|
2012-11-22 17:03:11 +00:00
|
|
|
if (type.width == 32 || type.length == 4) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vmaxfp";
|
|
|
|
intr_size = 128;
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
} else if (util_get_cpu_caps()->has_altivec) {
|
2012-11-22 17:03:11 +00:00
|
|
|
intr_size = 128;
|
|
|
|
if (type.width == 8) {
|
|
|
|
if (!type.sign) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vmaxub";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vmaxsb";
|
|
|
|
}
|
|
|
|
} else if (type.width == 16) {
|
|
|
|
if (!type.sign) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vmaxuh";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vmaxsh";
|
|
|
|
}
|
|
|
|
} else if (type.width == 32) {
|
|
|
|
if (!type.sign) {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vmaxuw";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.ppc.altivec.vmaxsw";
|
|
|
|
}
|
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (intrinsic) {
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse && type.floating &&
|
2021-04-29 21:55:30 +01:00
|
|
|
nan_behavior == GALLIVM_NAN_RETURN_OTHER) {
|
2014-08-30 03:58:58 +01:00
|
|
|
LLVMValueRef isnan, max;
|
|
|
|
max = lp_build_intrinsic_binary_anylength(bld->gallivm, intrinsic,
|
2013-07-16 18:06:24 +01:00
|
|
|
type,
|
|
|
|
intr_size, a, b);
|
2021-04-29 21:55:30 +01:00
|
|
|
isnan = lp_build_isnan(bld, b);
|
|
|
|
return lp_build_select(bld, isnan, a, max);
|
2013-07-16 18:06:24 +01:00
|
|
|
} else {
|
|
|
|
return lp_build_intrinsic_binary_anylength(bld->gallivm, intrinsic,
|
|
|
|
type,
|
|
|
|
intr_size, a, b);
|
|
|
|
}
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2013-07-16 18:06:24 +01:00
|
|
|
if (type.floating) {
|
|
|
|
switch (nan_behavior) {
|
|
|
|
case GALLIVM_NAN_RETURN_OTHER: {
|
|
|
|
LLVMValueRef isnan = lp_build_isnan(bld, a);
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b);
|
|
|
|
cond = LLVMBuildXor(bld->gallivm->builder, cond, isnan, "");
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
}
|
|
|
|
break;
|
2013-11-11 14:29:25 +00:00
|
|
|
case GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN:
|
2013-07-16 18:06:24 +01:00
|
|
|
cond = lp_build_cmp_ordered(bld, PIPE_FUNC_GREATER, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
2014-08-29 04:12:00 +01:00
|
|
|
case GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN:
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, b, a);
|
|
|
|
return lp_build_select(bld, cond, b, a);
|
2013-07-16 18:06:24 +01:00
|
|
|
case GALLIVM_NAN_BEHAVIOR_UNDEFINED:
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b);
|
|
|
|
return lp_build_select(bld, cond, a, b);
|
|
|
|
}
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate 1 - a, or ~a depending on bld->type.
|
|
|
|
*/
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef
|
2009-08-02 17:54:53 +01:00
|
|
|
lp_build_comp(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
2009-08-01 17:27:05 +01:00
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
if(a == bld->one)
|
|
|
|
return bld->zero;
|
|
|
|
if(a == bld->zero)
|
|
|
|
return bld->one;
|
|
|
|
|
|
|
|
if(type.norm && !type.floating && !type.fixed && !type.sign) {
|
|
|
|
if(LLVMIsConstant(a))
|
|
|
|
return LLVMConstNot(a);
|
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildNot(builder, a, "");
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if(LLVMIsConstant(a))
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
|
|
|
return LLVMConstFSub(bld->one, a);
|
|
|
|
else
|
|
|
|
return LLVMConstSub(bld->one, a);
|
2009-08-01 17:27:05 +01:00
|
|
|
else
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildFSub(builder, bld->one, a, "");
|
2010-08-08 20:44:54 +01:00
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildSub(builder, bld->one, a, "");
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate a + b
|
|
|
|
*/
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef
|
2009-08-02 17:54:53 +01:00
|
|
|
lp_build_add(struct lp_build_context *bld,
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-02 17:54:53 +01:00
|
|
|
LLVMValueRef res;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2010-08-08 21:02:59 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
assert(lp_check_value(type, b));
|
|
|
|
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (a == bld->zero)
|
2009-08-02 17:54:53 +01:00
|
|
|
return b;
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (b == bld->zero)
|
2009-08-02 17:54:53 +01:00
|
|
|
return a;
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (a == bld->undef || b == bld->undef)
|
2009-08-02 17:54:53 +01:00
|
|
|
return bld->undef;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (type.norm) {
|
2009-08-02 17:54:53 +01:00
|
|
|
const char *intrinsic = NULL;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (!type.sign && (a == bld->one || b == bld->one))
|
2009-08-02 17:54:53 +01:00
|
|
|
return bld->one;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2013-11-20 08:32:52 +00:00
|
|
|
if (!type.floating && !type.fixed) {
|
2019-10-17 03:14:28 +01:00
|
|
|
if (LLVM_VERSION_MAJOR >= 8) {
|
2019-04-17 01:34:01 +01:00
|
|
|
char intrin[32];
|
|
|
|
intrinsic = type.sign ? "llvm.sadd.sat" : "llvm.uadd.sat";
|
|
|
|
lp_format_intrinsic(intrin, sizeof intrin, intrinsic, bld->vec_type);
|
|
|
|
return lp_build_intrinsic_binary(builder, intrin, bld->vec_type, a, b);
|
|
|
|
}
|
2013-11-20 08:32:52 +00:00
|
|
|
if (type.width * type.length == 128) {
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse2) {
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 8)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.sse2.padds.b" : "llvm.x86.sse2.paddus.b";
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 16)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.sse2.padds.w" : "llvm.x86.sse2.paddus.w";
|
2021-02-25 18:04:50 +00:00
|
|
|
} else if (util_get_cpu_caps()->has_altivec) {
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 8)
|
|
|
|
intrinsic = type.sign ? "llvm.ppc.altivec.vaddsbs" : "llvm.ppc.altivec.vaddubs";
|
|
|
|
if (type.width == 16)
|
|
|
|
intrinsic = type.sign ? "llvm.ppc.altivec.vaddshs" : "llvm.ppc.altivec.vadduhs";
|
2013-11-20 08:32:52 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
if (type.width * type.length == 256) {
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_avx2) {
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 8)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.avx2.padds.b" : "llvm.x86.avx2.paddus.b";
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 16)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.avx2.padds.w" : "llvm.x86.avx2.paddus.w";
|
2013-11-20 08:32:52 +00:00
|
|
|
}
|
2012-11-22 17:20:42 +00:00
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (intrinsic)
|
2010-12-03 01:11:13 +00:00
|
|
|
return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b);
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
|
|
|
|
2014-07-21 16:53:34 +01:00
|
|
|
if(type.norm && !type.floating && !type.fixed) {
|
|
|
|
if (type.sign) {
|
|
|
|
uint64_t sign = (uint64_t)1 << (type.width - 1);
|
|
|
|
LLVMValueRef max_val = lp_build_const_int_vec(bld->gallivm, type, sign - 1);
|
|
|
|
LLVMValueRef min_val = lp_build_const_int_vec(bld->gallivm, type, sign);
|
|
|
|
/* a_clamp_max is the maximum a for positive b,
|
|
|
|
a_clamp_min is the minimum a for negative b. */
|
|
|
|
LLVMValueRef a_clamp_max = lp_build_min_simple(bld, a, LLVMBuildSub(builder, max_val, b, ""), GALLIVM_NAN_BEHAVIOR_UNDEFINED);
|
|
|
|
LLVMValueRef a_clamp_min = lp_build_max_simple(bld, a, LLVMBuildSub(builder, min_val, b, ""), GALLIVM_NAN_BEHAVIOR_UNDEFINED);
|
|
|
|
a = lp_build_select(bld, lp_build_cmp(bld, PIPE_FUNC_GREATER, b, bld->zero), a_clamp_max, a_clamp_min);
|
|
|
|
}
|
|
|
|
}
|
2013-06-07 13:31:58 +01:00
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
if(LLVMIsConstant(a) && LLVMIsConstant(b))
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
|
|
|
res = LLVMConstFAdd(a, b);
|
|
|
|
else
|
|
|
|
res = LLVMConstAdd(a, b);
|
2009-08-02 17:54:53 +01:00
|
|
|
else
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFAdd(builder, a, b, "");
|
2010-08-08 20:44:54 +01:00
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildAdd(builder, a, b, "");
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/* clamp to ceiling of 1.0 */
|
2009-08-02 17:54:53 +01:00
|
|
|
if(bld->type.norm && (bld->type.floating || bld->type.fixed))
|
2021-04-28 23:34:13 +01:00
|
|
|
res = lp_build_min_simple(bld, res, bld->one, GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN);
|
2009-08-02 17:54:53 +01:00
|
|
|
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.norm && !type.floating && !type.fixed) {
|
|
|
|
if (!type.sign) {
|
|
|
|
/*
|
|
|
|
* newer llvm versions no longer support the intrinsics, but recognize
|
|
|
|
* the pattern. Since auto-upgrade of intrinsics doesn't work for jit
|
|
|
|
* code, it is important we match the pattern llvm uses (and pray llvm
|
|
|
|
* doesn't change it - and hope they decide on the same pattern for
|
|
|
|
* all backends supporting it...).
|
|
|
|
* NOTE: cmp/select does sext/trunc of the mask. Does not seem to
|
|
|
|
* interfere with llvm's ability to recognize the pattern but seems
|
|
|
|
* a bit brittle.
|
2019-04-17 01:34:01 +01:00
|
|
|
* NOTE: llvm 9+ always uses (non arch specific) intrinsic.
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
*/
|
|
|
|
LLVMValueRef overflowed = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, res);
|
|
|
|
res = lp_build_select(bld, overflowed,
|
|
|
|
LLVMConstAllOnes(bld->int_vec_type), res);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/* XXX clamp to floor of -1 or 0??? */
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
return res;
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
/** Return the scalar sum of the elements of a.
|
|
|
|
* Should avoid this operation whenever possible.
|
|
|
|
*/
|
2010-03-11 23:26:12 +00:00
|
|
|
LLVMValueRef
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
lp_build_horizontal_add(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
2010-03-11 23:26:12 +00:00
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-03-11 23:26:12 +00:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
LLVMValueRef index, res;
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
unsigned i, length;
|
|
|
|
LLVMValueRef shuffles1[LP_MAX_VECTOR_LENGTH / 2];
|
|
|
|
LLVMValueRef shuffles2[LP_MAX_VECTOR_LENGTH / 2];
|
|
|
|
LLVMValueRef vecres, elem2;
|
2010-03-11 23:26:12 +00:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2010-08-21 11:31:22 +01:00
|
|
|
if (type.length == 1) {
|
|
|
|
return a;
|
|
|
|
}
|
2010-03-11 23:26:12 +00:00
|
|
|
|
|
|
|
assert(!bld->type.norm);
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
/*
|
|
|
|
* for byte vectors can do much better with psadbw.
|
|
|
|
* Using repeated shuffle/adds here. Note with multiple vectors
|
|
|
|
* this can be done more efficiently as outlined in the intel
|
|
|
|
* optimization manual.
|
|
|
|
* Note: could cause data rearrangement if used with smaller element
|
|
|
|
* sizes.
|
|
|
|
*/
|
2010-03-11 23:26:12 +00:00
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
vecres = a;
|
|
|
|
length = type.length / 2;
|
|
|
|
while (length > 1) {
|
|
|
|
LLVMValueRef vec1, vec2;
|
|
|
|
for (i = 0; i < length; i++) {
|
|
|
|
shuffles1[i] = lp_build_const_int32(bld->gallivm, i);
|
|
|
|
shuffles2[i] = lp_build_const_int32(bld->gallivm, i + length);
|
|
|
|
}
|
|
|
|
vec1 = LLVMBuildShuffleVector(builder, vecres, vecres,
|
|
|
|
LLVMConstVector(shuffles1, length), "");
|
|
|
|
vec2 = LLVMBuildShuffleVector(builder, vecres, vecres,
|
|
|
|
LLVMConstVector(shuffles2, length), "");
|
|
|
|
if (type.floating) {
|
|
|
|
vecres = LLVMBuildFAdd(builder, vec1, vec2, "");
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
vecres = LLVMBuildAdd(builder, vec1, vec2, "");
|
|
|
|
}
|
|
|
|
length = length >> 1;
|
2010-03-11 23:26:12 +00:00
|
|
|
}
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
/* always have vector of size 2 here */
|
|
|
|
assert(length == 1);
|
|
|
|
|
|
|
|
index = lp_build_const_int32(bld->gallivm, 0);
|
|
|
|
res = LLVMBuildExtractElement(builder, vecres, index, "");
|
|
|
|
index = lp_build_const_int32(bld->gallivm, 1);
|
|
|
|
elem2 = LLVMBuildExtractElement(builder, vecres, index, "");
|
|
|
|
|
|
|
|
if (type.floating)
|
|
|
|
res = LLVMBuildFAdd(builder, res, elem2, "");
|
|
|
|
else
|
|
|
|
res = LLVMBuildAdd(builder, res, elem2, "");
|
|
|
|
|
2010-03-11 23:26:12 +00:00
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
/**
|
|
|
|
* Return the horizontal sums of 4 float vectors as a float4 vector.
|
|
|
|
* This uses the technique as outlined in Intel Optimization Manual.
|
|
|
|
*/
|
|
|
|
static LLVMValueRef
|
|
|
|
lp_build_horizontal_add4x4f(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef src[4])
|
|
|
|
{
|
|
|
|
struct gallivm_state *gallivm = bld->gallivm;
|
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
|
|
|
LLVMValueRef shuffles[4];
|
|
|
|
LLVMValueRef tmp[4];
|
|
|
|
LLVMValueRef sumtmp[2], shuftmp[2];
|
|
|
|
|
|
|
|
/* lower half of regs */
|
|
|
|
shuffles[0] = lp_build_const_int32(gallivm, 0);
|
|
|
|
shuffles[1] = lp_build_const_int32(gallivm, 1);
|
|
|
|
shuffles[2] = lp_build_const_int32(gallivm, 4);
|
|
|
|
shuffles[3] = lp_build_const_int32(gallivm, 5);
|
|
|
|
tmp[0] = LLVMBuildShuffleVector(builder, src[0], src[1],
|
|
|
|
LLVMConstVector(shuffles, 4), "");
|
|
|
|
tmp[2] = LLVMBuildShuffleVector(builder, src[2], src[3],
|
|
|
|
LLVMConstVector(shuffles, 4), "");
|
|
|
|
|
|
|
|
/* upper half of regs */
|
|
|
|
shuffles[0] = lp_build_const_int32(gallivm, 2);
|
|
|
|
shuffles[1] = lp_build_const_int32(gallivm, 3);
|
|
|
|
shuffles[2] = lp_build_const_int32(gallivm, 6);
|
|
|
|
shuffles[3] = lp_build_const_int32(gallivm, 7);
|
|
|
|
tmp[1] = LLVMBuildShuffleVector(builder, src[0], src[1],
|
|
|
|
LLVMConstVector(shuffles, 4), "");
|
|
|
|
tmp[3] = LLVMBuildShuffleVector(builder, src[2], src[3],
|
|
|
|
LLVMConstVector(shuffles, 4), "");
|
|
|
|
|
|
|
|
sumtmp[0] = LLVMBuildFAdd(builder, tmp[0], tmp[1], "");
|
|
|
|
sumtmp[1] = LLVMBuildFAdd(builder, tmp[2], tmp[3], "");
|
|
|
|
|
|
|
|
shuffles[0] = lp_build_const_int32(gallivm, 0);
|
|
|
|
shuffles[1] = lp_build_const_int32(gallivm, 2);
|
|
|
|
shuffles[2] = lp_build_const_int32(gallivm, 4);
|
|
|
|
shuffles[3] = lp_build_const_int32(gallivm, 6);
|
|
|
|
shuftmp[0] = LLVMBuildShuffleVector(builder, sumtmp[0], sumtmp[1],
|
|
|
|
LLVMConstVector(shuffles, 4), "");
|
|
|
|
|
|
|
|
shuffles[0] = lp_build_const_int32(gallivm, 1);
|
|
|
|
shuffles[1] = lp_build_const_int32(gallivm, 3);
|
|
|
|
shuffles[2] = lp_build_const_int32(gallivm, 5);
|
|
|
|
shuffles[3] = lp_build_const_int32(gallivm, 7);
|
|
|
|
shuftmp[1] = LLVMBuildShuffleVector(builder, sumtmp[0], sumtmp[1],
|
|
|
|
LLVMConstVector(shuffles, 4), "");
|
|
|
|
|
|
|
|
return LLVMBuildFAdd(builder, shuftmp[0], shuftmp[1], "");
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* partially horizontally add 2-4 float vectors with length nx4,
|
|
|
|
* i.e. only four adjacent values in each vector will be added,
|
|
|
|
* assuming values are really grouped in 4 which also determines
|
|
|
|
* output order.
|
|
|
|
*
|
|
|
|
* Return a vector of the same length as the initial vectors,
|
|
|
|
* with the excess elements (if any) being undefined.
|
|
|
|
* The element order is independent of number of input vectors.
|
|
|
|
* For 3 vectors x0x1x2x3x4x5x6x7, y0y1y2y3y4y5y6y7, z0z1z2z3z4z5z6z7
|
|
|
|
* the output order thus will be
|
|
|
|
* sumx0-x3,sumy0-y3,sumz0-z3,undef,sumx4-x7,sumy4-y7,sumz4z7,undef
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_hadd_partial4(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef vectors[],
|
|
|
|
unsigned num_vecs)
|
|
|
|
{
|
|
|
|
struct gallivm_state *gallivm = bld->gallivm;
|
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
|
|
|
LLVMValueRef ret_vec;
|
|
|
|
LLVMValueRef tmp[4];
|
|
|
|
const char *intrinsic = NULL;
|
|
|
|
|
|
|
|
assert(num_vecs >= 2 && num_vecs <= 4);
|
|
|
|
assert(bld->type.floating);
|
|
|
|
|
|
|
|
/* only use this with at least 2 vectors, as it is sort of expensive
|
|
|
|
* (depending on cpu) and we always need two horizontal adds anyway,
|
|
|
|
* so a shuffle/add approach might be better.
|
|
|
|
*/
|
|
|
|
|
|
|
|
tmp[0] = vectors[0];
|
|
|
|
tmp[1] = vectors[1];
|
|
|
|
|
|
|
|
tmp[2] = num_vecs > 2 ? vectors[2] : vectors[0];
|
|
|
|
tmp[3] = num_vecs > 3 ? vectors[3] : vectors[0];
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse3 && bld->type.width == 32 &&
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
bld->type.length == 4) {
|
|
|
|
intrinsic = "llvm.x86.sse3.hadd.ps";
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (util_get_cpu_caps()->has_avx && bld->type.width == 32 &&
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
bld->type.length == 8) {
|
|
|
|
intrinsic = "llvm.x86.avx.hadd.ps.256";
|
|
|
|
}
|
|
|
|
if (intrinsic) {
|
|
|
|
tmp[0] = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
lp_build_vec_type(gallivm, bld->type),
|
|
|
|
tmp[0], tmp[1]);
|
|
|
|
if (num_vecs > 2) {
|
|
|
|
tmp[1] = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
lp_build_vec_type(gallivm, bld->type),
|
|
|
|
tmp[2], tmp[3]);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
tmp[1] = tmp[0];
|
|
|
|
}
|
|
|
|
return lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
lp_build_vec_type(gallivm, bld->type),
|
|
|
|
tmp[0], tmp[1]);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (bld->type.length == 4) {
|
|
|
|
ret_vec = lp_build_horizontal_add4x4f(bld, tmp);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
LLVMValueRef partres[LP_MAX_VECTOR_LENGTH/4];
|
|
|
|
unsigned j;
|
|
|
|
unsigned num_iter = bld->type.length / 4;
|
|
|
|
struct lp_type parttype = bld->type;
|
|
|
|
parttype.length = 4;
|
|
|
|
for (j = 0; j < num_iter; j++) {
|
|
|
|
LLVMValueRef partsrc[4];
|
|
|
|
unsigned i;
|
|
|
|
for (i = 0; i < 4; i++) {
|
|
|
|
partsrc[i] = lp_build_extract_range(gallivm, tmp[i], j*4, 4);
|
|
|
|
}
|
|
|
|
partres[j] = lp_build_horizontal_add4x4f(bld, partsrc);
|
|
|
|
}
|
|
|
|
ret_vec = lp_build_concat(gallivm, partres, parttype, num_iter);
|
|
|
|
}
|
|
|
|
return ret_vec;
|
|
|
|
}
|
2010-03-11 23:26:12 +00:00
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate a - b
|
|
|
|
*/
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef
|
2009-08-02 17:54:53 +01:00
|
|
|
lp_build_sub(struct lp_build_context *bld,
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-02 17:54:53 +01:00
|
|
|
LLVMValueRef res;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2010-08-08 21:02:59 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
assert(lp_check_value(type, b));
|
|
|
|
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (b == bld->zero)
|
2009-08-02 17:54:53 +01:00
|
|
|
return a;
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (a == bld->undef || b == bld->undef)
|
2009-08-02 17:54:53 +01:00
|
|
|
return bld->undef;
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (a == b)
|
2009-08-02 17:54:53 +01:00
|
|
|
return bld->zero;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (type.norm) {
|
2009-08-02 17:54:53 +01:00
|
|
|
const char *intrinsic = NULL;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
if (!type.sign && b == bld->one)
|
2009-08-02 17:54:53 +01:00
|
|
|
return bld->zero;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2013-11-20 08:32:52 +00:00
|
|
|
if (!type.floating && !type.fixed) {
|
2019-10-17 03:14:28 +01:00
|
|
|
if (LLVM_VERSION_MAJOR >= 8) {
|
2019-04-17 01:34:01 +01:00
|
|
|
char intrin[32];
|
|
|
|
intrinsic = type.sign ? "llvm.ssub.sat" : "llvm.usub.sat";
|
|
|
|
lp_format_intrinsic(intrin, sizeof intrin, intrinsic, bld->vec_type);
|
|
|
|
return lp_build_intrinsic_binary(builder, intrin, bld->vec_type, a, b);
|
|
|
|
}
|
2013-11-20 08:32:52 +00:00
|
|
|
if (type.width * type.length == 128) {
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse2) {
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 8)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.sse2.psubs.b" : "llvm.x86.sse2.psubus.b";
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 16)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.sse2.psubs.w" : "llvm.x86.sse2.psubus.w";
|
2021-02-25 18:04:50 +00:00
|
|
|
} else if (util_get_cpu_caps()->has_altivec) {
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 8)
|
|
|
|
intrinsic = type.sign ? "llvm.ppc.altivec.vsubsbs" : "llvm.ppc.altivec.vsububs";
|
|
|
|
if (type.width == 16)
|
|
|
|
intrinsic = type.sign ? "llvm.ppc.altivec.vsubshs" : "llvm.ppc.altivec.vsubuhs";
|
2013-11-20 08:32:52 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
if (type.width * type.length == 256) {
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_avx2) {
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 8)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.avx2.psubs.b" : "llvm.x86.avx2.psubus.b";
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
if (type.width == 16)
|
2019-10-17 03:14:28 +01:00
|
|
|
intrinsic = type.sign ? "llvm.x86.avx2.psubs.w" : "llvm.x86.avx2.psubus.w";
|
2013-11-20 08:32:52 +00:00
|
|
|
}
|
2012-11-22 17:20:42 +00:00
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (intrinsic)
|
2010-12-03 01:11:13 +00:00
|
|
|
return lp_build_intrinsic_binary(builder, intrinsic, lp_build_vec_type(bld->gallivm, bld->type), a, b);
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
|
|
|
|
2014-07-21 16:53:34 +01:00
|
|
|
if(type.norm && !type.floating && !type.fixed) {
|
|
|
|
if (type.sign) {
|
|
|
|
uint64_t sign = (uint64_t)1 << (type.width - 1);
|
|
|
|
LLVMValueRef max_val = lp_build_const_int_vec(bld->gallivm, type, sign - 1);
|
|
|
|
LLVMValueRef min_val = lp_build_const_int_vec(bld->gallivm, type, sign);
|
|
|
|
/* a_clamp_max is the maximum a for negative b,
|
|
|
|
a_clamp_min is the minimum a for positive b. */
|
|
|
|
LLVMValueRef a_clamp_max = lp_build_min_simple(bld, a, LLVMBuildAdd(builder, max_val, b, ""), GALLIVM_NAN_BEHAVIOR_UNDEFINED);
|
|
|
|
LLVMValueRef a_clamp_min = lp_build_max_simple(bld, a, LLVMBuildAdd(builder, min_val, b, ""), GALLIVM_NAN_BEHAVIOR_UNDEFINED);
|
|
|
|
a = lp_build_select(bld, lp_build_cmp(bld, PIPE_FUNC_GREATER, b, bld->zero), a_clamp_min, a_clamp_max);
|
|
|
|
} else {
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
/*
|
|
|
|
* This must match llvm pattern for saturated unsigned sub.
|
|
|
|
* (lp_build_max_simple actually does the job with its current
|
|
|
|
* definition but do it explicitly here.)
|
|
|
|
* NOTE: cmp/select does sext/trunc of the mask. Does not seem to
|
|
|
|
* interfere with llvm's ability to recognize the pattern but seems
|
|
|
|
* a bit brittle.
|
2019-04-17 01:34:01 +01:00
|
|
|
* NOTE: llvm 9+ always uses (non arch specific) intrinsic.
|
gallivm: don't use saturated unsigned add/sub intrinsics for llvm 8.0
These have been removed. Unfortunately auto-upgrade doesn't work for
jit. (Worse, it seems we don't get a compilation error anymore when
compiling the shader, rather llvm will just do a call to a null
function in the jitted shaders making it difficult to detect when
intrinsics vanish.)
Luckily the signed ones are still there, I helped convincing llvm
removing them is a bad idea for now, since while the unsigned ones have
sort of agreed-upon simplest patterns to replace them with, this is not
the case for the signed ones, and they require _significantly_ more
complex patterns - to the point that the recognition is IMHO probably
unlikely to ever work reliably in practice (due to other optimizations
interfering). (Even for the relatively trivial unsigned patterns, llvm
already added test cases where recognition doesn't work, unsaturated
add followed by saturated add may produce atrocious code.)
Nevertheless, it seems there's a serious quest to squash all
cpu-specific intrinsics going on, so I'd expect patches to nuke them as
well to resurface.
Adapt the existing fallback code to match the simple patterns llvm uses
and hope for the best. I've verified with lp_test_blend that it does
produce the expected saturated assembly instructions. Though our
cmp/select build helpers don't use boolean masks, but it doesn't seem
to interfere with llvm's ability to recognize the pattern.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106231
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2018-08-23 18:07:05 +01:00
|
|
|
*/
|
|
|
|
LLVMValueRef no_ov = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, b);
|
|
|
|
a = lp_build_select(bld, no_ov, a, b);
|
2014-07-21 16:53:34 +01:00
|
|
|
}
|
|
|
|
}
|
2013-06-07 13:31:58 +01:00
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
if(LLVMIsConstant(a) && LLVMIsConstant(b))
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
|
|
|
res = LLVMConstFSub(a, b);
|
|
|
|
else
|
|
|
|
res = LLVMConstSub(a, b);
|
2009-08-02 17:54:53 +01:00
|
|
|
else
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFSub(builder, a, b, "");
|
2010-08-08 20:44:54 +01:00
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildSub(builder, a, b, "");
|
2009-08-02 17:54:53 +01:00
|
|
|
|
|
|
|
if(bld->type.norm && (bld->type.floating || bld->type.fixed))
|
2021-04-28 23:34:13 +01:00
|
|
|
res = lp_build_max_simple(bld, res, bld->zero, GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN);
|
2009-08-02 17:54:53 +01:00
|
|
|
|
|
|
|
return res;
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2012-12-04 13:38:52 +00:00
|
|
|
|
2009-08-03 00:01:27 +01:00
|
|
|
/**
|
2012-12-04 13:38:52 +00:00
|
|
|
* Normalized multiplication.
|
|
|
|
*
|
2012-12-06 09:30:53 +00:00
|
|
|
* There are several approaches for (using 8-bit normalized multiplication as
|
2012-12-04 13:38:52 +00:00
|
|
|
* an example):
|
2009-08-03 00:01:27 +01:00
|
|
|
*
|
|
|
|
* - alpha plus one
|
|
|
|
*
|
|
|
|
* makes the following approximation to the division (Sree)
|
|
|
|
*
|
|
|
|
* a*b/255 ~= (a*(b + 1)) >> 256
|
|
|
|
*
|
2012-12-06 09:30:53 +00:00
|
|
|
* which is the fastest method that satisfies the following OpenGL criteria of
|
2009-08-03 00:01:27 +01:00
|
|
|
*
|
|
|
|
* 0*0 = 0 and 255*255 = 255
|
|
|
|
*
|
|
|
|
* - geometric series
|
|
|
|
*
|
|
|
|
* takes the geometric series approximation to the division
|
|
|
|
*
|
|
|
|
* t/255 = (t >> 8) + (t >> 16) + (t >> 24) ..
|
|
|
|
*
|
|
|
|
* in this case just the first two terms to fit in 16bit arithmetic
|
|
|
|
*
|
|
|
|
* t/255 ~= (t + (t >> 8)) >> 8
|
|
|
|
*
|
|
|
|
* note that just by itself it doesn't satisfies the OpenGL criteria, as
|
|
|
|
* 255*255 = 254, so the special case b = 255 must be accounted or roundoff
|
2012-12-06 09:30:53 +00:00
|
|
|
* must be used.
|
2009-08-03 00:01:27 +01:00
|
|
|
*
|
|
|
|
* - geometric series plus rounding
|
|
|
|
*
|
|
|
|
* when using a geometric series division instead of truncating the result
|
|
|
|
* use roundoff in the approximation (Jim Blinn)
|
|
|
|
*
|
|
|
|
* t/255 ~= (t + (t >> 8) + 0x80) >> 8
|
|
|
|
*
|
2012-12-06 09:30:53 +00:00
|
|
|
* achieving the exact results.
|
|
|
|
*
|
|
|
|
*
|
2009-08-03 00:01:27 +01:00
|
|
|
*
|
|
|
|
* @sa Alvy Ray Smith, Image Compositing Fundamentals, Tech Memo 4, Aug 15, 1995,
|
|
|
|
* ftp://ftp.alvyray.com/Acrobat/4_Comp.pdf
|
|
|
|
* @sa Michael Herf, The "double blend trick", May 2000,
|
|
|
|
* http://www.stereopsis.com/doubleblend.html
|
|
|
|
*/
|
llvmpipe: fix snorm blending
The blend math gets a bit funky due to inverse blend factors being
in range [0,2] rather than [-1,1], our normalized math can't really
cover this.
src_alpha_saturate blend factor has a similar problem too.
(Note that piglit fbo-blending-formats test is mostly useless for
anything but unorm formats, since not just all src/dst values are
between [0,1], but the tests are crafted in a way that the results
are between [0,1] too.)
v2: some formatting fixes, and fix a fairly obscure (to debug)
issue with alpha-only formats (not related to snorm at all), where
blend optimization would think it could simplify the blend equation
if the blend factors were complementary, however was using the
completely unrelated rgb blend factors instead of the alpha ones...
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2017-11-18 05:23:35 +00:00
|
|
|
LLVMValueRef
|
2012-12-04 13:38:52 +00:00
|
|
|
lp_build_mul_norm(struct gallivm_state *gallivm,
|
|
|
|
struct lp_type wide_type,
|
|
|
|
LLVMValueRef a, LLVMValueRef b)
|
2009-08-03 00:01:27 +01:00
|
|
|
{
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
2012-12-04 13:38:52 +00:00
|
|
|
struct lp_build_context bld;
|
2012-12-06 09:30:53 +00:00
|
|
|
unsigned n;
|
2012-12-04 13:38:52 +00:00
|
|
|
LLVMValueRef half;
|
2009-08-03 00:01:27 +01:00
|
|
|
LLVMValueRef ab;
|
|
|
|
|
2012-12-04 13:38:52 +00:00
|
|
|
assert(!wide_type.floating);
|
|
|
|
assert(lp_check_value(wide_type, a));
|
|
|
|
assert(lp_check_value(wide_type, b));
|
|
|
|
|
|
|
|
lp_build_context_init(&bld, gallivm, wide_type);
|
|
|
|
|
2012-12-06 09:30:53 +00:00
|
|
|
n = wide_type.width / 2;
|
2012-12-04 13:38:52 +00:00
|
|
|
if (wide_type.sign) {
|
2012-12-06 09:30:53 +00:00
|
|
|
--n;
|
2012-12-04 13:38:52 +00:00
|
|
|
}
|
|
|
|
|
2012-12-06 09:30:53 +00:00
|
|
|
/*
|
|
|
|
* TODO: for 16bits normalized SSE2 vectors we could consider using PMULHUW
|
|
|
|
* http://ssp.impulsetrain.com/2011/07/03/multiplying-normalized-16-bit-numbers-with-sse2/
|
|
|
|
*/
|
2010-08-09 22:15:08 +01:00
|
|
|
|
2012-12-06 09:30:53 +00:00
|
|
|
/*
|
|
|
|
* a*b / (2**n - 1) ~= (a*b + (a*b >> n) + half) >> n
|
|
|
|
*/
|
2009-08-03 00:01:27 +01:00
|
|
|
|
|
|
|
ab = LLVMBuildMul(builder, a, b, "");
|
2012-12-06 09:30:53 +00:00
|
|
|
ab = LLVMBuildAdd(builder, ab, lp_build_shr_imm(&bld, ab, n), "");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* half = sgn(ab) * 0.5 * (2 ** n) = sgn(ab) * (1 << (n - 1))
|
|
|
|
*/
|
2012-12-04 13:38:52 +00:00
|
|
|
|
2014-11-25 23:03:02 +00:00
|
|
|
half = lp_build_const_int_vec(gallivm, wide_type, 1LL << (n - 1));
|
2012-12-04 13:38:52 +00:00
|
|
|
if (wide_type.sign) {
|
|
|
|
LLVMValueRef minus_half = LLVMBuildNeg(builder, half, "");
|
2013-02-01 16:34:36 +00:00
|
|
|
LLVMValueRef sign = lp_build_shr_imm(&bld, ab, wide_type.width - 1);
|
2012-12-04 13:38:52 +00:00
|
|
|
half = lp_build_select(&bld, sign, minus_half, half);
|
|
|
|
}
|
|
|
|
ab = LLVMBuildAdd(builder, ab, half, "");
|
2009-08-03 00:01:27 +01:00
|
|
|
|
2012-12-06 09:30:53 +00:00
|
|
|
/* Final division */
|
|
|
|
ab = lp_build_shr_imm(&bld, ab, n);
|
2012-05-30 14:36:22 +01:00
|
|
|
|
|
|
|
return ab;
|
|
|
|
}
|
2009-08-03 00:01:27 +01:00
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate a * b
|
|
|
|
*/
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef
|
2009-08-02 17:54:53 +01:00
|
|
|
lp_build_mul(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b)
|
2009-08-01 17:27:05 +01:00
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-10-25 09:16:38 +00:00
|
|
|
LLVMValueRef shift;
|
|
|
|
LLVMValueRef res;
|
2009-08-03 00:01:27 +01:00
|
|
|
|
2010-08-08 21:02:59 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
assert(lp_check_value(type, b));
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
if(a == bld->zero)
|
|
|
|
return bld->zero;
|
|
|
|
if(a == bld->one)
|
2009-08-01 17:27:05 +01:00
|
|
|
return b;
|
2009-08-02 17:54:53 +01:00
|
|
|
if(b == bld->zero)
|
|
|
|
return bld->zero;
|
|
|
|
if(b == bld->one)
|
2009-08-01 17:27:05 +01:00
|
|
|
return a;
|
2009-08-02 17:54:53 +01:00
|
|
|
if(a == bld->undef || b == bld->undef)
|
|
|
|
return bld->undef;
|
2009-08-01 17:27:05 +01:00
|
|
|
|
2012-12-04 13:38:52 +00:00
|
|
|
if (!type.floating && !type.fixed && type.norm) {
|
|
|
|
struct lp_type wide_type = lp_wider_type(type);
|
|
|
|
LLVMValueRef al, ah, bl, bh, abl, abh, ab;
|
2009-08-03 00:01:27 +01:00
|
|
|
|
gallivm: Use native packs and unpacks for the lerps
For the texturing packs, things looked pretty terrible. For every
lerp, we were repacking the values, and while those look sort of cheap
with 128bit, with 256bit we end up with 2 of them instead of just 1 but
worse, plus 2 extracts too (the unpack, however, works fine with a
single instruction, albeit only with llvm 3.8 - the vpmovzxbw).
Ideally we'd use more clever pack for llvmpipe backend conversion too
since we actually use the "wrong" shuffle (which is more work) when doing
the fs twiddle just so we end up with the wrong order for being able to
do native pack when converting from 2x8f -> 1x16b. But this requires some
refactoring, since the untwiddle is separate from conversion.
This is only used for avx2 256bit pack/unpack for now.
Improves openarena scores by 8% or so, though overall it's still pretty
disappointing how much faster 256bit vectors are even with avx2 (or
rather, aren't...). And, of course, eliminating the needless
packs/unpacks in the first place would eliminate most of that advantage
(not quite all) from this patch.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2016-10-18 02:37:37 +01:00
|
|
|
lp_build_unpack2_native(bld->gallivm, type, wide_type, a, &al, &ah);
|
|
|
|
lp_build_unpack2_native(bld->gallivm, type, wide_type, b, &bl, &bh);
|
2012-05-30 14:36:22 +01:00
|
|
|
|
2012-12-04 13:38:52 +00:00
|
|
|
/* PMULLW, PSRLW, PADDW */
|
|
|
|
abl = lp_build_mul_norm(bld->gallivm, wide_type, al, bl);
|
|
|
|
abh = lp_build_mul_norm(bld->gallivm, wide_type, ah, bh);
|
2012-05-30 14:36:22 +01:00
|
|
|
|
gallivm: Use native packs and unpacks for the lerps
For the texturing packs, things looked pretty terrible. For every
lerp, we were repacking the values, and while those look sort of cheap
with 128bit, with 256bit we end up with 2 of them instead of just 1 but
worse, plus 2 extracts too (the unpack, however, works fine with a
single instruction, albeit only with llvm 3.8 - the vpmovzxbw).
Ideally we'd use more clever pack for llvmpipe backend conversion too
since we actually use the "wrong" shuffle (which is more work) when doing
the fs twiddle just so we end up with the wrong order for being able to
do native pack when converting from 2x8f -> 1x16b. But this requires some
refactoring, since the untwiddle is separate from conversion.
This is only used for avx2 256bit pack/unpack for now.
Improves openarena scores by 8% or so, though overall it's still pretty
disappointing how much faster 256bit vectors are even with avx2 (or
rather, aren't...). And, of course, eliminating the needless
packs/unpacks in the first place would eliminate most of that advantage
(not quite all) from this patch.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2016-10-18 02:37:37 +01:00
|
|
|
ab = lp_build_pack2_native(bld->gallivm, wide_type, type, abl, abh);
|
2012-05-30 14:36:22 +01:00
|
|
|
|
2012-12-04 13:38:52 +00:00
|
|
|
return ab;
|
2009-08-03 00:01:27 +01:00
|
|
|
}
|
|
|
|
|
2009-10-25 09:16:38 +00:00
|
|
|
if(type.fixed)
|
2010-11-30 23:07:52 +00:00
|
|
|
shift = lp_build_const_int_vec(bld->gallivm, type, type.width/2);
|
2009-10-25 09:16:38 +00:00
|
|
|
else
|
|
|
|
shift = NULL;
|
|
|
|
|
|
|
|
if(LLVMIsConstant(a) && LLVMIsConstant(b)) {
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
|
|
|
res = LLVMConstFMul(a, b);
|
|
|
|
else
|
|
|
|
res = LLVMConstMul(a, b);
|
2009-10-25 09:16:38 +00:00
|
|
|
if(shift) {
|
|
|
|
if(type.sign)
|
|
|
|
res = LLVMConstAShr(res, shift);
|
|
|
|
else
|
|
|
|
res = LLVMConstLShr(res, shift);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else {
|
2010-08-08 20:44:54 +01:00
|
|
|
if (type.floating)
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFMul(builder, a, b, "");
|
2010-08-08 20:44:54 +01:00
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildMul(builder, a, b, "");
|
2009-10-25 09:16:38 +00:00
|
|
|
if(shift) {
|
|
|
|
if(type.sign)
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildAShr(builder, res, shift, "");
|
2009-10-25 09:16:38 +00:00
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildLShr(builder, res, shift, "");
|
2009-10-25 09:16:38 +00:00
|
|
|
}
|
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2009-10-25 09:16:38 +00:00
|
|
|
return res;
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
2016-11-04 03:55:09 +00:00
|
|
|
/*
|
|
|
|
* Widening mul, valid for 32x32 bit -> 64bit only.
|
|
|
|
* Result is low 32bits, high bits returned in res_hi.
|
2016-11-08 09:14:00 +00:00
|
|
|
*
|
|
|
|
* Emits code that is meant to be compiled for the host CPU.
|
2016-11-04 03:55:09 +00:00
|
|
|
*/
|
|
|
|
LLVMValueRef
|
2016-11-08 09:14:00 +00:00
|
|
|
lp_build_mul_32_lohi_cpu(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b,
|
|
|
|
LLVMValueRef *res_hi)
|
2016-11-04 03:55:09 +00:00
|
|
|
{
|
|
|
|
struct gallivm_state *gallivm = bld->gallivm;
|
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
|
|
|
|
|
|
|
assert(bld->type.width == 32);
|
|
|
|
assert(bld->type.floating == 0);
|
|
|
|
assert(bld->type.fixed == 0);
|
|
|
|
assert(bld->type.norm == 0);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* XXX: for some reason, with zext/zext/mul/trunc the code llvm produces
|
|
|
|
* for x86 simd is atrocious (even if the high bits weren't required),
|
|
|
|
* trying to handle real 64bit inputs (which of course can't happen due
|
|
|
|
* to using 64bit umul with 32bit numbers zero-extended to 64bit, but
|
|
|
|
* apparently llvm does not recognize this widening mul). This includes 6
|
|
|
|
* (instead of 2) pmuludq plus extra adds and shifts
|
|
|
|
* The same story applies to signed mul, albeit fixing this requires sse41.
|
|
|
|
* https://llvm.org/bugs/show_bug.cgi?id=30845
|
|
|
|
* So, whip up our own code, albeit only for length 4 and 8 (which
|
|
|
|
* should be good enough)...
|
2019-08-28 20:35:45 +01:00
|
|
|
* FIXME: For llvm >= 7.0 we should match the autoupgrade pattern
|
|
|
|
* (bitcast/and/mul/shuffle for unsigned, bitcast/shl/ashr/mul/shuffle
|
|
|
|
* for signed), which the fallback code does not, without this llvm
|
|
|
|
* will likely still produce atrocious code.
|
2016-11-04 03:55:09 +00:00
|
|
|
*/
|
2019-08-28 00:06:21 +01:00
|
|
|
if (LLVM_VERSION_MAJOR < 7 &&
|
2019-08-28 20:35:45 +01:00
|
|
|
(bld->type.length == 4 || bld->type.length == 8) &&
|
2021-02-25 18:04:50 +00:00
|
|
|
((util_get_cpu_caps()->has_sse2 && (bld->type.sign == 0)) ||
|
|
|
|
util_get_cpu_caps()->has_sse4_1)) {
|
2016-11-04 03:55:09 +00:00
|
|
|
const char *intrinsic = NULL;
|
|
|
|
LLVMValueRef aeven, aodd, beven, bodd, muleven, mulodd;
|
|
|
|
LLVMValueRef shuf[LP_MAX_VECTOR_WIDTH / 32], shuf_vec;
|
|
|
|
struct lp_type type_wide = lp_wider_type(bld->type);
|
|
|
|
LLVMTypeRef wider_type = lp_build_vec_type(gallivm, type_wide);
|
|
|
|
unsigned i;
|
|
|
|
for (i = 0; i < bld->type.length; i += 2) {
|
|
|
|
shuf[i] = lp_build_const_int32(gallivm, i+1);
|
|
|
|
shuf[i+1] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context));
|
|
|
|
}
|
|
|
|
shuf_vec = LLVMConstVector(shuf, bld->type.length);
|
|
|
|
aeven = a;
|
|
|
|
beven = b;
|
|
|
|
aodd = LLVMBuildShuffleVector(builder, aeven, bld->undef, shuf_vec, "");
|
|
|
|
bodd = LLVMBuildShuffleVector(builder, beven, bld->undef, shuf_vec, "");
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_avx2 && bld->type.length == 8) {
|
2016-11-04 03:55:09 +00:00
|
|
|
if (bld->type.sign) {
|
|
|
|
intrinsic = "llvm.x86.avx2.pmul.dq";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.x86.avx2.pmulu.dq";
|
|
|
|
}
|
|
|
|
muleven = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wider_type, aeven, beven);
|
|
|
|
mulodd = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wider_type, aodd, bodd);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
/* for consistent naming look elsewhere... */
|
|
|
|
if (bld->type.sign) {
|
|
|
|
intrinsic = "llvm.x86.sse41.pmuldq";
|
|
|
|
} else {
|
|
|
|
intrinsic = "llvm.x86.sse2.pmulu.dq";
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* XXX If we only have AVX but not AVX2 this is a pain.
|
|
|
|
* lp_build_intrinsic_binary_anylength() can't handle it
|
|
|
|
* (due to src and dst type not being identical).
|
|
|
|
*/
|
|
|
|
if (bld->type.length == 8) {
|
|
|
|
LLVMValueRef aevenlo, aevenhi, bevenlo, bevenhi;
|
|
|
|
LLVMValueRef aoddlo, aoddhi, boddlo, boddhi;
|
|
|
|
LLVMValueRef muleven2[2], mulodd2[2];
|
|
|
|
struct lp_type type_wide_half = type_wide;
|
|
|
|
LLVMTypeRef wtype_half;
|
|
|
|
type_wide_half.length = 2;
|
|
|
|
wtype_half = lp_build_vec_type(gallivm, type_wide_half);
|
|
|
|
aevenlo = lp_build_extract_range(gallivm, aeven, 0, 4);
|
|
|
|
aevenhi = lp_build_extract_range(gallivm, aeven, 4, 4);
|
|
|
|
bevenlo = lp_build_extract_range(gallivm, beven, 0, 4);
|
|
|
|
bevenhi = lp_build_extract_range(gallivm, beven, 4, 4);
|
|
|
|
aoddlo = lp_build_extract_range(gallivm, aodd, 0, 4);
|
|
|
|
aoddhi = lp_build_extract_range(gallivm, aodd, 4, 4);
|
|
|
|
boddlo = lp_build_extract_range(gallivm, bodd, 0, 4);
|
|
|
|
boddhi = lp_build_extract_range(gallivm, bodd, 4, 4);
|
|
|
|
muleven2[0] = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wtype_half, aevenlo, bevenlo);
|
|
|
|
mulodd2[0] = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wtype_half, aoddlo, boddlo);
|
|
|
|
muleven2[1] = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wtype_half, aevenhi, bevenhi);
|
|
|
|
mulodd2[1] = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wtype_half, aoddhi, boddhi);
|
|
|
|
muleven = lp_build_concat(gallivm, muleven2, type_wide_half, 2);
|
|
|
|
mulodd = lp_build_concat(gallivm, mulodd2, type_wide_half, 2);
|
|
|
|
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
muleven = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wider_type, aeven, beven);
|
|
|
|
mulodd = lp_build_intrinsic_binary(builder, intrinsic,
|
|
|
|
wider_type, aodd, bodd);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
muleven = LLVMBuildBitCast(builder, muleven, bld->vec_type, "");
|
|
|
|
mulodd = LLVMBuildBitCast(builder, mulodd, bld->vec_type, "");
|
|
|
|
|
|
|
|
for (i = 0; i < bld->type.length; i += 2) {
|
|
|
|
shuf[i] = lp_build_const_int32(gallivm, i + 1);
|
|
|
|
shuf[i+1] = lp_build_const_int32(gallivm, i + 1 + bld->type.length);
|
|
|
|
}
|
|
|
|
shuf_vec = LLVMConstVector(shuf, bld->type.length);
|
|
|
|
*res_hi = LLVMBuildShuffleVector(builder, muleven, mulodd, shuf_vec, "");
|
|
|
|
|
|
|
|
for (i = 0; i < bld->type.length; i += 2) {
|
|
|
|
shuf[i] = lp_build_const_int32(gallivm, i);
|
|
|
|
shuf[i+1] = lp_build_const_int32(gallivm, i + bld->type.length);
|
|
|
|
}
|
|
|
|
shuf_vec = LLVMConstVector(shuf, bld->type.length);
|
|
|
|
return LLVMBuildShuffleVector(builder, muleven, mulodd, shuf_vec, "");
|
|
|
|
}
|
|
|
|
else {
|
2016-11-08 09:14:00 +00:00
|
|
|
return lp_build_mul_32_lohi(bld, a, b, res_hi);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
2020-10-08 22:13:45 +01:00
|
|
|
* Widening mul, valid for <= 32 (8, 16, 32) -> 64
|
|
|
|
* Result is low N bits, high bits returned in res_hi.
|
2016-11-08 09:14:00 +00:00
|
|
|
*
|
|
|
|
* Emits generic code.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_mul_32_lohi(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b,
|
|
|
|
LLVMValueRef *res_hi)
|
|
|
|
{
|
|
|
|
struct gallivm_state *gallivm = bld->gallivm;
|
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
2016-11-09 11:44:47 +00:00
|
|
|
LLVMValueRef tmp, shift, res_lo;
|
2016-11-08 09:14:00 +00:00
|
|
|
struct lp_type type_tmp;
|
2016-11-09 11:44:47 +00:00
|
|
|
LLVMTypeRef wide_type, narrow_type;
|
2016-11-08 09:14:00 +00:00
|
|
|
|
|
|
|
type_tmp = bld->type;
|
2016-11-09 11:44:47 +00:00
|
|
|
narrow_type = lp_build_vec_type(gallivm, type_tmp);
|
2020-10-08 22:13:45 +01:00
|
|
|
if (bld->type.width < 32)
|
|
|
|
type_tmp.width = 32;
|
|
|
|
else
|
|
|
|
type_tmp.width *= 2;
|
2016-11-08 09:14:00 +00:00
|
|
|
wide_type = lp_build_vec_type(gallivm, type_tmp);
|
2020-10-08 22:13:45 +01:00
|
|
|
shift = lp_build_const_vec(gallivm, type_tmp, bld->type.width);
|
2016-11-08 09:14:00 +00:00
|
|
|
|
|
|
|
if (bld->type.sign) {
|
|
|
|
a = LLVMBuildSExt(builder, a, wide_type, "");
|
|
|
|
b = LLVMBuildSExt(builder, b, wide_type, "");
|
|
|
|
} else {
|
|
|
|
a = LLVMBuildZExt(builder, a, wide_type, "");
|
|
|
|
b = LLVMBuildZExt(builder, b, wide_type, "");
|
2016-11-04 03:55:09 +00:00
|
|
|
}
|
2016-11-08 09:14:00 +00:00
|
|
|
tmp = LLVMBuildMul(builder, a, b, "");
|
2016-11-09 11:44:47 +00:00
|
|
|
|
|
|
|
res_lo = LLVMBuildTrunc(builder, tmp, narrow_type, "");
|
|
|
|
|
|
|
|
/* Since we truncate anyway, LShr and AShr are equivalent. */
|
|
|
|
tmp = LLVMBuildLShr(builder, tmp, shift, "");
|
|
|
|
*res_hi = LLVMBuildTrunc(builder, tmp, narrow_type, "");
|
|
|
|
|
|
|
|
return res_lo;
|
2016-11-04 03:55:09 +00:00
|
|
|
}
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2016-04-04 00:05:33 +01:00
|
|
|
/* a * b + c */
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_mad(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b,
|
|
|
|
LLVMValueRef c)
|
|
|
|
{
|
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
if (type.floating) {
|
|
|
|
return lp_build_fmuladd(bld->gallivm->builder, a, b, c);
|
|
|
|
} else {
|
|
|
|
return lp_build_add(bld, lp_build_mul(bld, a, b), c);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-10-25 11:48:17 +00:00
|
|
|
/**
|
|
|
|
* Small vector x scale multiplication optimization.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_mul_imm(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
int b)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-10-25 11:48:17 +00:00
|
|
|
LLVMValueRef factor;
|
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, a));
|
|
|
|
|
2009-10-25 11:48:17 +00:00
|
|
|
if(b == 0)
|
|
|
|
return bld->zero;
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2009-10-25 11:48:17 +00:00
|
|
|
if(b == 1)
|
|
|
|
return a;
|
|
|
|
|
|
|
|
if(b == -1)
|
2010-08-09 17:26:18 +01:00
|
|
|
return lp_build_negate(bld, a);
|
2009-10-25 11:48:17 +00:00
|
|
|
|
|
|
|
if(b == 2 && bld->type.floating)
|
|
|
|
return lp_build_add(bld, a, a);
|
|
|
|
|
2017-11-13 19:17:41 +00:00
|
|
|
if(util_is_power_of_two_or_zero(b)) {
|
2009-10-25 11:48:17 +00:00
|
|
|
unsigned shift = ffs(b) - 1;
|
|
|
|
|
|
|
|
if(bld->type.floating) {
|
|
|
|
#if 0
|
|
|
|
/*
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
* Power of two multiplication by directly manipulating the exponent.
|
2009-10-25 11:48:17 +00:00
|
|
|
*
|
|
|
|
* XXX: This might not be always faster, it will introduce a small error
|
|
|
|
* for multiplication by zero, and it will produce wrong results
|
|
|
|
* for Inf and NaN.
|
|
|
|
*/
|
|
|
|
unsigned mantissa = lp_mantissa(bld->type);
|
2010-11-30 23:07:52 +00:00
|
|
|
factor = lp_build_const_int_vec(bld->gallivm, bld->type, (unsigned long long)shift << mantissa);
|
2010-12-03 01:11:13 +00:00
|
|
|
a = LLVMBuildBitCast(builder, a, lp_build_int_vec_type(bld->type), "");
|
|
|
|
a = LLVMBuildAdd(builder, a, factor, "");
|
|
|
|
a = LLVMBuildBitCast(builder, a, lp_build_vec_type(bld->gallivm, bld->type), "");
|
2009-10-25 11:48:17 +00:00
|
|
|
return a;
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
else {
|
2010-11-30 23:07:52 +00:00
|
|
|
factor = lp_build_const_vec(bld->gallivm, bld->type, shift);
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildShl(builder, a, factor, "");
|
2009-10-25 11:48:17 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-11-30 23:07:52 +00:00
|
|
|
factor = lp_build_const_vec(bld->gallivm, bld->type, (double)b);
|
2009-10-25 11:48:17 +00:00
|
|
|
return lp_build_mul(bld, a, factor);
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate a / b
|
|
|
|
*/
|
2009-08-11 13:09:35 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_div(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-08-08 21:02:59 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
assert(lp_check_value(type, b));
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
if(a == bld->zero)
|
|
|
|
return bld->zero;
|
2014-09-08 12:56:13 +01:00
|
|
|
if(a == bld->one && type.floating)
|
2009-08-11 13:09:35 +01:00
|
|
|
return lp_build_rcp(bld, b);
|
|
|
|
if(b == bld->zero)
|
|
|
|
return bld->undef;
|
|
|
|
if(b == bld->one)
|
|
|
|
return a;
|
|
|
|
if(a == bld->undef || b == bld->undef)
|
|
|
|
return bld->undef;
|
|
|
|
|
2010-08-09 22:25:18 +01:00
|
|
|
if(LLVMIsConstant(a) && LLVMIsConstant(b)) {
|
|
|
|
if (type.floating)
|
|
|
|
return LLVMConstFDiv(a, b);
|
|
|
|
else if (type.sign)
|
|
|
|
return LLVMConstSDiv(a, b);
|
|
|
|
else
|
|
|
|
return LLVMConstUDiv(a, b);
|
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2017-01-23 17:06:03 +00:00
|
|
|
/* fast rcp is disabled (just uses div), so makes no sense to try that */
|
|
|
|
if(FALSE &&
|
2021-02-25 18:04:50 +00:00
|
|
|
((util_get_cpu_caps()->has_sse && type.width == 32 && type.length == 4) ||
|
|
|
|
(util_get_cpu_caps()->has_avx && type.width == 32 && type.length == 8)) &&
|
2012-02-28 11:23:04 +00:00
|
|
|
type.floating)
|
2009-08-11 13:09:35 +01:00
|
|
|
return lp_build_mul(bld, a, lp_build_rcp(bld, b));
|
|
|
|
|
2010-08-09 22:25:18 +01:00
|
|
|
if (type.floating)
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildFDiv(builder, a, b, "");
|
2010-08-09 22:25:18 +01:00
|
|
|
else if (type.sign)
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildSDiv(builder, a, b, "");
|
2010-08-09 22:25:18 +01:00
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildUDiv(builder, a, b, "");
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-10-25 09:16:38 +00:00
|
|
|
/**
|
2012-12-04 13:38:52 +00:00
|
|
|
* Linear interpolation helper.
|
|
|
|
*
|
|
|
|
* @param normalized whether we are interpolating normalized values,
|
|
|
|
* encoded in normalized integers, twice as wide.
|
2009-10-25 09:16:38 +00:00
|
|
|
*
|
|
|
|
* @sa http://www.stereopsis.com/doubleblend.html
|
|
|
|
*/
|
2015-07-21 00:58:43 +01:00
|
|
|
static inline LLVMValueRef
|
2010-09-11 13:21:44 +01:00
|
|
|
lp_build_lerp_simple(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
LLVMValueRef v0,
|
2012-12-04 13:38:52 +00:00
|
|
|
LLVMValueRef v1,
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
unsigned flags)
|
2009-09-07 14:27:06 +01:00
|
|
|
{
|
2012-12-04 13:38:52 +00:00
|
|
|
unsigned half_width = bld->type.width/2;
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-10-25 09:16:38 +00:00
|
|
|
LLVMValueRef delta;
|
|
|
|
LLVMValueRef res;
|
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
assert(lp_check_value(bld->type, v0));
|
|
|
|
assert(lp_check_value(bld->type, v1));
|
|
|
|
|
2009-10-25 09:16:38 +00:00
|
|
|
delta = lp_build_sub(bld, v1, v0);
|
|
|
|
|
2016-04-04 00:05:33 +01:00
|
|
|
if (bld->type.floating) {
|
|
|
|
assert(flags == 0);
|
|
|
|
return lp_build_mad(bld, x, delta, v0);
|
|
|
|
}
|
|
|
|
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
if (flags & LP_BLD_LERP_WIDE_NORMALIZED) {
|
2012-12-06 09:30:53 +00:00
|
|
|
if (!bld->type.sign) {
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
if (!(flags & LP_BLD_LERP_PRESCALED_WEIGHTS)) {
|
|
|
|
/*
|
|
|
|
* Scale x from [0, 2**n - 1] to [0, 2**n] by adding the
|
|
|
|
* most-significant-bit to the lowest-significant-bit, so that
|
|
|
|
* later we can just divide by 2**n instead of 2**n - 1.
|
|
|
|
*/
|
|
|
|
|
|
|
|
x = lp_build_add(bld, x, lp_build_shr_imm(bld, x, half_width - 1));
|
|
|
|
}
|
2012-12-06 09:30:53 +00:00
|
|
|
|
|
|
|
/* (x * delta) >> n */
|
2021-09-28 05:39:37 +01:00
|
|
|
/*
|
|
|
|
* For this multiply, higher internal precision is required to pass CTS,
|
|
|
|
* the most efficient path to that is pmulhrsw on ssse3 and above.
|
|
|
|
* This could be opencoded on other arches if conformance was required.
|
|
|
|
*/
|
|
|
|
if (bld->type.width == 16 && bld->type.length == 8 && util_get_cpu_caps()->has_ssse3) {
|
|
|
|
res = lp_build_intrinsic_binary(builder, "llvm.x86.ssse3.pmul.hr.sw.128", bld->vec_type, x, lp_build_shl_imm(bld, delta, 7));
|
|
|
|
res = lp_build_and(bld, res, lp_build_const_int_vec(bld->gallivm, bld->type, 0xff));
|
|
|
|
} else if (bld->type.width == 16 && bld->type.length == 16 && util_get_cpu_caps()->has_avx2) {
|
|
|
|
res = lp_build_intrinsic_binary(builder, "llvm.x86.avx2.pmul.hr.sw", bld->vec_type, x, lp_build_shl_imm(bld, delta, 7));
|
|
|
|
res = lp_build_and(bld, res, lp_build_const_int_vec(bld->gallivm, bld->type, 0xff));
|
|
|
|
} else {
|
|
|
|
res = lp_build_mul(bld, x, delta);
|
|
|
|
res = lp_build_shr_imm(bld, res, half_width);
|
|
|
|
}
|
2012-12-06 09:30:53 +00:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* The rescaling trick above doesn't work for signed numbers, so
|
|
|
|
* use the 2**n - 1 divison approximation in lp_build_mul_norm
|
|
|
|
* instead.
|
|
|
|
*/
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
assert(!(flags & LP_BLD_LERP_PRESCALED_WEIGHTS));
|
2012-12-06 09:30:53 +00:00
|
|
|
res = lp_build_mul_norm(bld->gallivm, bld->type, x, delta);
|
2012-12-04 13:38:52 +00:00
|
|
|
}
|
2012-12-06 09:30:53 +00:00
|
|
|
} else {
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
assert(!(flags & LP_BLD_LERP_PRESCALED_WEIGHTS));
|
2012-12-06 09:30:53 +00:00
|
|
|
res = lp_build_mul(bld, x, delta);
|
2012-12-04 13:38:52 +00:00
|
|
|
}
|
|
|
|
|
2016-05-13 00:44:39 +01:00
|
|
|
if ((flags & LP_BLD_LERP_WIDE_NORMALIZED) && !bld->type.sign) {
|
|
|
|
/*
|
|
|
|
* At this point both res and v0 only use the lower half of the bits,
|
|
|
|
* the rest is zero. Instead of add / mask, do add with half wide type.
|
|
|
|
*/
|
|
|
|
struct lp_type narrow_type;
|
|
|
|
struct lp_build_context narrow_bld;
|
|
|
|
|
|
|
|
memset(&narrow_type, 0, sizeof narrow_type);
|
|
|
|
narrow_type.sign = bld->type.sign;
|
|
|
|
narrow_type.width = bld->type.width/2;
|
|
|
|
narrow_type.length = bld->type.length*2;
|
|
|
|
|
|
|
|
lp_build_context_init(&narrow_bld, bld->gallivm, narrow_type);
|
|
|
|
res = LLVMBuildBitCast(builder, res, narrow_bld.vec_type, "");
|
|
|
|
v0 = LLVMBuildBitCast(builder, v0, narrow_bld.vec_type, "");
|
|
|
|
res = lp_build_add(&narrow_bld, v0, res);
|
|
|
|
res = LLVMBuildBitCast(builder, res, bld->vec_type, "");
|
|
|
|
} else {
|
|
|
|
res = lp_build_add(bld, v0, res);
|
|
|
|
|
|
|
|
if (bld->type.fixed) {
|
|
|
|
/*
|
|
|
|
* We need to mask out the high order bits when lerping 8bit
|
|
|
|
* normalized colors stored on 16bits
|
|
|
|
*/
|
|
|
|
/* XXX: This step is necessary for lerping 8bit colors stored on
|
|
|
|
* 16bits, but it will be wrong for true fixed point use cases.
|
|
|
|
* Basically we need a more powerful lp_type, capable of further
|
|
|
|
* distinguishing the values interpretation from the value storage.
|
|
|
|
*/
|
|
|
|
LLVMValueRef low_bits;
|
|
|
|
low_bits = lp_build_const_int_vec(bld->gallivm, bld->type, (1 << half_width) - 1);
|
|
|
|
res = LLVMBuildAnd(builder, res, low_bits, "");
|
|
|
|
}
|
2010-09-11 13:21:44 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Linear interpolation.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_lerp(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
LLVMValueRef v0,
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
LLVMValueRef v1,
|
|
|
|
unsigned flags)
|
2010-09-11 13:21:44 +01:00
|
|
|
{
|
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
LLVMValueRef res;
|
|
|
|
|
|
|
|
assert(lp_check_value(type, x));
|
|
|
|
assert(lp_check_value(type, v0));
|
|
|
|
assert(lp_check_value(type, v1));
|
|
|
|
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
assert(!(flags & LP_BLD_LERP_WIDE_NORMALIZED));
|
|
|
|
|
2010-09-11 13:21:44 +01:00
|
|
|
if (type.norm) {
|
|
|
|
struct lp_type wide_type;
|
|
|
|
struct lp_build_context wide_bld;
|
|
|
|
LLVMValueRef xl, xh, v0l, v0h, v1l, v1h, resl, resh;
|
|
|
|
|
|
|
|
assert(type.length >= 2);
|
|
|
|
|
|
|
|
/*
|
2012-12-04 13:38:52 +00:00
|
|
|
* Create a wider integer type, enough to hold the
|
|
|
|
* intermediate result of the multiplication.
|
2010-09-11 13:21:44 +01:00
|
|
|
*/
|
|
|
|
memset(&wide_type, 0, sizeof wide_type);
|
2012-12-04 13:38:52 +00:00
|
|
|
wide_type.sign = type.sign;
|
2010-09-11 13:21:44 +01:00
|
|
|
wide_type.width = type.width*2;
|
|
|
|
wide_type.length = type.length/2;
|
|
|
|
|
2010-11-30 23:07:52 +00:00
|
|
|
lp_build_context_init(&wide_bld, bld->gallivm, wide_type);
|
2010-09-11 13:21:44 +01:00
|
|
|
|
gallivm: Use native packs and unpacks for the lerps
For the texturing packs, things looked pretty terrible. For every
lerp, we were repacking the values, and while those look sort of cheap
with 128bit, with 256bit we end up with 2 of them instead of just 1 but
worse, plus 2 extracts too (the unpack, however, works fine with a
single instruction, albeit only with llvm 3.8 - the vpmovzxbw).
Ideally we'd use more clever pack for llvmpipe backend conversion too
since we actually use the "wrong" shuffle (which is more work) when doing
the fs twiddle just so we end up with the wrong order for being able to
do native pack when converting from 2x8f -> 1x16b. But this requires some
refactoring, since the untwiddle is separate from conversion.
This is only used for avx2 256bit pack/unpack for now.
Improves openarena scores by 8% or so, though overall it's still pretty
disappointing how much faster 256bit vectors are even with avx2 (or
rather, aren't...). And, of course, eliminating the needless
packs/unpacks in the first place would eliminate most of that advantage
(not quite all) from this patch.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2016-10-18 02:37:37 +01:00
|
|
|
lp_build_unpack2_native(bld->gallivm, type, wide_type, x, &xl, &xh);
|
|
|
|
lp_build_unpack2_native(bld->gallivm, type, wide_type, v0, &v0l, &v0h);
|
|
|
|
lp_build_unpack2_native(bld->gallivm, type, wide_type, v1, &v1l, &v1h);
|
2010-09-11 13:21:44 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Lerp both halves.
|
|
|
|
*/
|
|
|
|
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
flags |= LP_BLD_LERP_WIDE_NORMALIZED;
|
|
|
|
|
|
|
|
resl = lp_build_lerp_simple(&wide_bld, xl, v0l, v1l, flags);
|
|
|
|
resh = lp_build_lerp_simple(&wide_bld, xh, v0h, v1h, flags);
|
2010-09-11 13:21:44 +01:00
|
|
|
|
gallivm: Use native packs and unpacks for the lerps
For the texturing packs, things looked pretty terrible. For every
lerp, we were repacking the values, and while those look sort of cheap
with 128bit, with 256bit we end up with 2 of them instead of just 1 but
worse, plus 2 extracts too (the unpack, however, works fine with a
single instruction, albeit only with llvm 3.8 - the vpmovzxbw).
Ideally we'd use more clever pack for llvmpipe backend conversion too
since we actually use the "wrong" shuffle (which is more work) when doing
the fs twiddle just so we end up with the wrong order for being able to
do native pack when converting from 2x8f -> 1x16b. But this requires some
refactoring, since the untwiddle is separate from conversion.
This is only used for avx2 256bit pack/unpack for now.
Improves openarena scores by 8% or so, though overall it's still pretty
disappointing how much faster 256bit vectors are even with avx2 (or
rather, aren't...). And, of course, eliminating the needless
packs/unpacks in the first place would eliminate most of that advantage
(not quite all) from this patch.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2016-10-18 02:37:37 +01:00
|
|
|
res = lp_build_pack2_native(bld->gallivm, wide_type, type, resl, resh);
|
2010-09-11 13:21:44 +01:00
|
|
|
} else {
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
res = lp_build_lerp_simple(bld, x, v0, v1, flags);
|
2010-09-11 13:21:44 +01:00
|
|
|
}
|
2009-10-25 09:16:38 +00:00
|
|
|
|
|
|
|
return res;
|
2009-09-07 14:27:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
/**
|
|
|
|
* Bilinear interpolation.
|
|
|
|
*
|
|
|
|
* Values indices are in v_{yx}.
|
|
|
|
*/
|
2009-09-07 14:27:06 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_lerp_2d(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
LLVMValueRef y,
|
|
|
|
LLVMValueRef v00,
|
|
|
|
LLVMValueRef v01,
|
|
|
|
LLVMValueRef v10,
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
LLVMValueRef v11,
|
|
|
|
unsigned flags)
|
2009-09-07 14:27:06 +01:00
|
|
|
{
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
LLVMValueRef v0 = lp_build_lerp(bld, x, v00, v01, flags);
|
|
|
|
LLVMValueRef v1 = lp_build_lerp(bld, x, v10, v11, flags);
|
|
|
|
return lp_build_lerp(bld, y, v0, v1, flags);
|
2009-09-07 14:27:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2013-04-17 18:03:11 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_lerp_3d(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
LLVMValueRef y,
|
|
|
|
LLVMValueRef z,
|
|
|
|
LLVMValueRef v000,
|
|
|
|
LLVMValueRef v001,
|
|
|
|
LLVMValueRef v010,
|
|
|
|
LLVMValueRef v011,
|
|
|
|
LLVMValueRef v100,
|
|
|
|
LLVMValueRef v101,
|
|
|
|
LLVMValueRef v110,
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
LLVMValueRef v111,
|
|
|
|
unsigned flags)
|
2013-04-17 18:03:11 +01:00
|
|
|
{
|
gallivm: Eliminate 8.8 fixed point intermediates from AoS sampling path.
This change was meant as a stepping stone to use PMADDUBSW SSSE3
instruction, but actually this refactoring by itself yields a 10%
speedup on texture intensive shaders (e.g, Google Earth's ocean water
w/o S3TC on a Ivy Bridge machine), while giving yielding exactly the
same results, whereas PMADDUBSW only gave an extra 5%, at the expense of
2bits of precision in the interpolation.
I belive that the speedup of this change comes from the reduced register
pressure (as 8.8 fixed point intermediates take twice the space of 8bit
unorm).
Also, not dealing with 8.8 simplifies lp_bld_sample_aos.c code
substantially -- it's no longer necessary to have code duplicated for
low and high register halfs.
Note about lp_build_sample_mipmap(): the path for num_quads > 1 is never
executed (as it is faster on AVX to split the 256bit wide texture
computation into two 128bit chunks, in order to leverage integer
opcodes). This path might be useful in the future, so in order to
verify this change did not break that path I had to apply this change:
@@ -1662,11 +1662,11 @@ lp_build_sample_soa(struct gallivm_state *gallivm,
/*
* we only try 8-wide sampling with soa as it appears to
* be a loss with aos with AVX (but it should work).
* (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || !use_aos) {
+ if (/* num_quads == 1 || ! */ use_aos) {
if (num_quads > 1) {
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
/*
and then run texfilt mesademo:
LP_NATIVE_VECTOR_WIDTH=256 ./texfilt
Ran whole piglit without regressions.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2013-04-21 22:23:31 +01:00
|
|
|
LLVMValueRef v0 = lp_build_lerp_2d(bld, x, y, v000, v001, v010, v011, flags);
|
|
|
|
LLVMValueRef v1 = lp_build_lerp_2d(bld, x, y, v100, v101, v110, v111, flags);
|
|
|
|
return lp_build_lerp(bld, z, v0, v1, flags);
|
2013-04-17 18:03:11 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate min(a, b)
|
2013-07-16 18:06:24 +01:00
|
|
|
* Do checks for special cases but not for nans.
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef
|
2009-08-02 17:54:53 +01:00
|
|
|
lp_build_min(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b)
|
2009-08-01 17:27:05 +01:00
|
|
|
{
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, a));
|
|
|
|
assert(lp_check_value(bld->type, b));
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
if(a == bld->undef || b == bld->undef)
|
|
|
|
return bld->undef;
|
|
|
|
|
2009-08-03 19:31:56 +01:00
|
|
|
if(a == b)
|
|
|
|
return a;
|
|
|
|
|
2012-05-18 16:06:44 +01:00
|
|
|
if (bld->type.norm) {
|
|
|
|
if (!bld->type.sign) {
|
|
|
|
if (a == bld->zero || b == bld->zero) {
|
|
|
|
return bld->zero;
|
|
|
|
}
|
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
if(a == bld->one)
|
|
|
|
return b;
|
|
|
|
if(b == bld->one)
|
|
|
|
return a;
|
|
|
|
}
|
|
|
|
|
2013-07-16 18:06:24 +01:00
|
|
|
return lp_build_min_simple(bld, a, b, GALLIVM_NAN_BEHAVIOR_UNDEFINED);
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
|
|
|
|
2013-07-16 18:06:24 +01:00
|
|
|
/**
|
|
|
|
* Generate min(a, b)
|
|
|
|
* NaN's are handled according to the behavior specified by the
|
|
|
|
* nan_behavior argument.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_min_ext(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b,
|
|
|
|
enum gallivm_nan_behavior nan_behavior)
|
|
|
|
{
|
|
|
|
assert(lp_check_value(bld->type, a));
|
|
|
|
assert(lp_check_value(bld->type, b));
|
|
|
|
|
|
|
|
if(a == bld->undef || b == bld->undef)
|
|
|
|
return bld->undef;
|
|
|
|
|
|
|
|
if(a == b)
|
|
|
|
return a;
|
|
|
|
|
|
|
|
if (bld->type.norm) {
|
|
|
|
if (!bld->type.sign) {
|
|
|
|
if (a == bld->zero || b == bld->zero) {
|
|
|
|
return bld->zero;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if(a == bld->one)
|
|
|
|
return b;
|
|
|
|
if(b == bld->one)
|
|
|
|
return a;
|
|
|
|
}
|
|
|
|
|
|
|
|
return lp_build_min_simple(bld, a, b, nan_behavior);
|
|
|
|
}
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate max(a, b)
|
2013-07-16 18:06:24 +01:00
|
|
|
* Do checks for special cases, but NaN behavior is undefined.
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2009-08-01 17:27:05 +01:00
|
|
|
LLVMValueRef
|
2009-08-02 17:54:53 +01:00
|
|
|
lp_build_max(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b)
|
2009-08-01 17:27:05 +01:00
|
|
|
{
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, a));
|
|
|
|
assert(lp_check_value(bld->type, b));
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
if(a == bld->undef || b == bld->undef)
|
|
|
|
return bld->undef;
|
|
|
|
|
2009-08-03 19:31:56 +01:00
|
|
|
if(a == b)
|
|
|
|
return a;
|
|
|
|
|
2009-08-02 17:54:53 +01:00
|
|
|
if(bld->type.norm) {
|
|
|
|
if(a == bld->one || b == bld->one)
|
|
|
|
return bld->one;
|
2012-05-18 16:06:44 +01:00
|
|
|
if (!bld->type.sign) {
|
|
|
|
if (a == bld->zero) {
|
|
|
|
return b;
|
|
|
|
}
|
|
|
|
if (b == bld->zero) {
|
|
|
|
return a;
|
|
|
|
}
|
|
|
|
}
|
2009-08-02 17:54:53 +01:00
|
|
|
}
|
|
|
|
|
2013-07-16 18:06:24 +01:00
|
|
|
return lp_build_max_simple(bld, a, b, GALLIVM_NAN_BEHAVIOR_UNDEFINED);
|
2009-08-01 17:27:05 +01:00
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
|
|
|
|
|
2013-07-16 18:06:24 +01:00
|
|
|
/**
|
|
|
|
* Generate max(a, b)
|
|
|
|
* Checks for special cases.
|
|
|
|
* NaN's are handled according to the behavior specified by the
|
|
|
|
* nan_behavior argument.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_max_ext(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef b,
|
|
|
|
enum gallivm_nan_behavior nan_behavior)
|
|
|
|
{
|
|
|
|
assert(lp_check_value(bld->type, a));
|
|
|
|
assert(lp_check_value(bld->type, b));
|
|
|
|
|
|
|
|
if(a == bld->undef || b == bld->undef)
|
|
|
|
return bld->undef;
|
|
|
|
|
|
|
|
if(a == b)
|
|
|
|
return a;
|
|
|
|
|
|
|
|
if(bld->type.norm) {
|
|
|
|
if(a == bld->one || b == bld->one)
|
|
|
|
return bld->one;
|
|
|
|
if (!bld->type.sign) {
|
|
|
|
if (a == bld->zero) {
|
|
|
|
return b;
|
|
|
|
}
|
|
|
|
if (b == bld->zero) {
|
|
|
|
return a;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return lp_build_max_simple(bld, a, b, nan_behavior);
|
|
|
|
}
|
|
|
|
|
2010-02-24 04:16:18 +00:00
|
|
|
/**
|
|
|
|
* Generate clamp(a, min, max)
|
2013-11-11 14:29:25 +00:00
|
|
|
* NaN behavior (for any of a, min, max) is undefined.
|
2010-02-24 04:16:18 +00:00
|
|
|
* Do checks for special cases.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_clamp(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef min,
|
|
|
|
LLVMValueRef max)
|
|
|
|
{
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, a));
|
|
|
|
assert(lp_check_value(bld->type, min));
|
|
|
|
assert(lp_check_value(bld->type, max));
|
|
|
|
|
2013-08-15 17:40:32 +01:00
|
|
|
a = lp_build_min(bld, a, max);
|
2013-08-15 18:26:39 +01:00
|
|
|
a = lp_build_max(bld, a, min);
|
2010-02-24 04:16:18 +00:00
|
|
|
return a;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2013-11-11 14:29:25 +00:00
|
|
|
/**
|
|
|
|
* Generate clamp(a, 0, 1)
|
|
|
|
* A NaN will get converted to zero.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_clamp_zero_one_nanzero(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
|
|
|
a = lp_build_max_ext(bld, a, bld->zero, GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN);
|
|
|
|
a = lp_build_min(bld, a, bld->one);
|
|
|
|
return a;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate abs(a)
|
|
|
|
*/
|
2009-08-11 13:09:35 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_abs(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
if(!type.sign)
|
|
|
|
return a;
|
|
|
|
|
2009-09-07 14:53:15 +01:00
|
|
|
if(type.floating) {
|
2019-09-06 08:50:32 +01:00
|
|
|
char intrinsic[32];
|
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.fabs", vec_type);
|
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a);
|
2009-09-07 14:53:15 +01:00
|
|
|
}
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if(type.width*type.length == 128 && util_get_cpu_caps()->has_ssse3 && LLVM_VERSION_MAJOR < 6) {
|
2009-09-07 14:53:15 +01:00
|
|
|
switch(type.width) {
|
|
|
|
case 8:
|
2010-12-03 01:11:13 +00:00
|
|
|
return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a);
|
2009-09-07 14:53:15 +01:00
|
|
|
case 16:
|
2010-12-03 01:11:13 +00:00
|
|
|
return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.w.128", vec_type, a);
|
2009-09-07 14:53:15 +01:00
|
|
|
case 32:
|
2010-12-03 01:11:13 +00:00
|
|
|
return lp_build_intrinsic_unary(builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a);
|
2009-09-07 14:53:15 +01:00
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (type.width*type.length == 256 && util_get_cpu_caps()->has_avx2 && LLVM_VERSION_MAJOR < 6) {
|
2013-11-20 08:32:52 +00:00
|
|
|
switch(type.width) {
|
|
|
|
case 8:
|
|
|
|
return lp_build_intrinsic_unary(builder, "llvm.x86.avx2.pabs.b", vec_type, a);
|
|
|
|
case 16:
|
|
|
|
return lp_build_intrinsic_unary(builder, "llvm.x86.avx2.pabs.w", vec_type, a);
|
|
|
|
case 32:
|
|
|
|
return lp_build_intrinsic_unary(builder, "llvm.x86.avx2.pabs.d", vec_type, a);
|
|
|
|
}
|
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2017-10-06 23:52:58 +01:00
|
|
|
return lp_build_select(bld, lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero),
|
|
|
|
a, LLVMBuildNeg(builder, a, ""));
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-03-04 15:36:26 +00:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_negate(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, a));
|
|
|
|
|
2010-08-08 20:44:54 +01:00
|
|
|
if (bld->type.floating)
|
2010-12-03 01:11:13 +00:00
|
|
|
a = LLVMBuildFNeg(builder, a, "");
|
2010-08-08 20:44:54 +01:00
|
|
|
else
|
2010-12-03 01:11:13 +00:00
|
|
|
a = LLVMBuildNeg(builder, a, "");
|
2010-08-08 20:44:54 +01:00
|
|
|
|
|
|
|
return a;
|
2010-03-04 15:36:26 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-03-12 21:24:03 +00:00
|
|
|
/** Return -1, 0 or +1 depending on the sign of a */
|
2009-09-13 13:50:19 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_sgn(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-09-13 13:50:19 +01:00
|
|
|
LLVMValueRef cond;
|
|
|
|
LLVMValueRef res;
|
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2009-09-13 13:50:19 +01:00
|
|
|
/* Handle non-zero case */
|
|
|
|
if(!type.sign) {
|
|
|
|
/* if not zero then sign must be positive */
|
|
|
|
res = bld->one;
|
|
|
|
}
|
|
|
|
else if(type.floating) {
|
2010-03-12 21:24:03 +00:00
|
|
|
LLVMTypeRef vec_type;
|
|
|
|
LLVMTypeRef int_type;
|
|
|
|
LLVMValueRef mask;
|
2009-09-13 13:50:19 +01:00
|
|
|
LLVMValueRef sign;
|
|
|
|
LLVMValueRef one;
|
2010-03-12 21:24:03 +00:00
|
|
|
unsigned long long maskBit = (unsigned long long)1 << (type.width - 1);
|
|
|
|
|
2010-11-30 23:07:52 +00:00
|
|
|
int_type = lp_build_int_vec_type(bld->gallivm, type);
|
|
|
|
vec_type = lp_build_vec_type(bld->gallivm, type);
|
|
|
|
mask = lp_build_const_int_vec(bld->gallivm, type, maskBit);
|
2010-03-12 21:24:03 +00:00
|
|
|
|
|
|
|
/* Take the sign bit and add it to 1 constant */
|
2010-12-03 01:11:13 +00:00
|
|
|
sign = LLVMBuildBitCast(builder, a, int_type, "");
|
|
|
|
sign = LLVMBuildAnd(builder, sign, mask, "");
|
2010-03-12 21:24:03 +00:00
|
|
|
one = LLVMConstBitCast(bld->one, int_type);
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildOr(builder, sign, one, "");
|
|
|
|
res = LLVMBuildBitCast(builder, res, vec_type, "");
|
2009-09-13 13:50:19 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2012-05-15 22:36:09 +01:00
|
|
|
/* signed int/norm/fixed point */
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
/* could use psign with sse3 and appropriate vectors here */
|
2012-05-15 22:36:09 +01:00
|
|
|
LLVMValueRef minus_one = lp_build_const_vec(bld->gallivm, type, -1.0);
|
2009-09-13 13:50:19 +01:00
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero);
|
|
|
|
res = lp_build_select(bld, cond, bld->one, minus_one);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Handle zero */
|
|
|
|
cond = lp_build_cmp(bld, PIPE_FUNC_EQUAL, a, bld->zero);
|
2010-03-15 17:27:24 +00:00
|
|
|
res = lp_build_select(bld, cond, bld->zero, res);
|
2009-09-13 13:50:19 +01:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-03-04 16:45:34 +00:00
|
|
|
/**
|
|
|
|
* Set the sign of float vector 'a' according to 'sign'.
|
|
|
|
* If sign==0, return abs(a).
|
|
|
|
* If sign==1, return -abs(a);
|
|
|
|
* Other values for sign produce undefined results.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_set_sign(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a, LLVMValueRef sign)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-03-04 16:45:34 +00:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
|
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
|
|
|
LLVMValueRef shift = lp_build_const_int_vec(bld->gallivm, type, type.width - 1);
|
|
|
|
LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type,
|
2010-03-04 16:45:34 +00:00
|
|
|
~((unsigned long long) 1 << (type.width - 1)));
|
|
|
|
LLVMValueRef val, res;
|
|
|
|
|
|
|
|
assert(type.floating);
|
2010-08-09 22:15:08 +01:00
|
|
|
assert(lp_check_value(type, a));
|
2010-03-04 16:45:34 +00:00
|
|
|
|
|
|
|
/* val = reinterpret_cast<int>(a) */
|
2010-12-03 01:11:13 +00:00
|
|
|
val = LLVMBuildBitCast(builder, a, int_vec_type, "");
|
2010-03-04 16:45:34 +00:00
|
|
|
/* val = val & mask */
|
2010-12-03 01:11:13 +00:00
|
|
|
val = LLVMBuildAnd(builder, val, mask, "");
|
2010-03-04 16:45:34 +00:00
|
|
|
/* sign = sign << shift */
|
2010-12-03 01:11:13 +00:00
|
|
|
sign = LLVMBuildShl(builder, sign, shift, "");
|
2010-03-04 16:45:34 +00:00
|
|
|
/* res = val | sign */
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildOr(builder, val, sign, "");
|
2010-03-04 16:45:34 +00:00
|
|
|
/* res = reinterpret_cast<float>(res) */
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildBitCast(builder, res, vec_type, "");
|
2010-03-04 16:45:34 +00:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-02-24 04:16:18 +00:00
|
|
|
/**
|
2010-03-09 16:37:50 +00:00
|
|
|
* Convert vector of (or scalar) int to vector of (or scalar) float.
|
2010-02-24 04:16:18 +00:00
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_int_to_float(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-02-24 04:16:18 +00:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
2010-02-24 04:16:18 +00:00
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildSIToFP(builder, a, vec_type, "");
|
2010-02-24 04:16:18 +00:00
|
|
|
}
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
static boolean
|
2012-11-22 17:37:18 +00:00
|
|
|
arch_rounding_available(const struct lp_type type)
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
{
|
2021-02-25 18:04:50 +00:00
|
|
|
if ((util_get_cpu_caps()->has_sse4_1 &&
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
(type.length == 1 || type.width*type.length == 128)) ||
|
2021-02-25 18:04:50 +00:00
|
|
|
(util_get_cpu_caps()->has_avx && type.width*type.length == 256) ||
|
|
|
|
(util_get_cpu_caps()->has_avx512f && type.width*type.length == 512))
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
return TRUE;
|
2021-02-25 18:04:50 +00:00
|
|
|
else if ((util_get_cpu_caps()->has_altivec &&
|
2012-11-22 17:37:18 +00:00
|
|
|
(type.width == 32 && type.length == 4)))
|
|
|
|
return TRUE;
|
2021-02-25 18:04:50 +00:00
|
|
|
else if (util_get_cpu_caps()->has_neon)
|
2019-01-22 23:06:38 +00:00
|
|
|
return TRUE;
|
2021-11-18 17:27:35 +00:00
|
|
|
else if (util_cpu_caps_has_zarch())
|
|
|
|
return TRUE;
|
2010-02-24 04:16:18 +00:00
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
return FALSE;
|
|
|
|
}
|
2010-02-24 04:16:18 +00:00
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
enum lp_build_round_mode
|
2009-09-07 14:26:30 +01:00
|
|
|
{
|
2012-11-22 17:37:18 +00:00
|
|
|
LP_BUILD_ROUND_NEAREST = 0,
|
|
|
|
LP_BUILD_ROUND_FLOOR = 1,
|
|
|
|
LP_BUILD_ROUND_CEIL = 2,
|
|
|
|
LP_BUILD_ROUND_TRUNCATE = 3
|
2009-09-07 14:26:30 +01:00
|
|
|
};
|
|
|
|
|
2015-07-21 00:58:43 +01:00
|
|
|
static inline LLVMValueRef
|
2010-10-08 17:38:25 +01:00
|
|
|
lp_build_iround_nearest_sse2(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-10-08 17:38:25 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef i32t = LLVMInt32TypeInContext(bld->gallivm->context);
|
|
|
|
LLVMTypeRef ret_type = lp_build_int_vec_type(bld->gallivm, type);
|
2010-10-08 17:38:25 +01:00
|
|
|
const char *intrinsic;
|
|
|
|
LLVMValueRef res;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
/* using the double precision conversions is a bit more complicated */
|
|
|
|
assert(type.width == 32);
|
|
|
|
|
|
|
|
assert(lp_check_value(type, a));
|
2021-02-25 18:04:50 +00:00
|
|
|
assert(util_get_cpu_caps()->has_sse2);
|
2010-10-08 17:38:25 +01:00
|
|
|
|
|
|
|
/* This is relying on MXCSR rounding mode, which should always be nearest. */
|
|
|
|
if (type.length == 1) {
|
|
|
|
LLVMTypeRef vec_type;
|
|
|
|
LLVMValueRef undef;
|
|
|
|
LLVMValueRef arg;
|
|
|
|
LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
|
|
|
|
|
|
|
|
vec_type = LLVMVectorType(bld->elem_type, 4);
|
|
|
|
|
|
|
|
intrinsic = "llvm.x86.sse.cvtss2si";
|
|
|
|
|
|
|
|
undef = LLVMGetUndef(vec_type);
|
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
arg = LLVMBuildInsertElement(builder, undef, a, index0, "");
|
2010-10-08 17:38:25 +01:00
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
res = lp_build_intrinsic_unary(builder, intrinsic,
|
2010-10-08 17:38:25 +01:00
|
|
|
ret_type, arg);
|
|
|
|
}
|
|
|
|
else {
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
if (type.width* type.length == 128) {
|
|
|
|
intrinsic = "llvm.x86.sse2.cvtps2dq";
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
assert(type.width*type.length == 256);
|
2021-02-25 18:04:50 +00:00
|
|
|
assert(util_get_cpu_caps()->has_avx);
|
2010-10-08 17:38:25 +01:00
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
intrinsic = "llvm.x86.avx.cvt.ps2dq.256";
|
|
|
|
}
|
2010-12-03 01:11:13 +00:00
|
|
|
res = lp_build_intrinsic_unary(builder, intrinsic,
|
2010-10-08 17:38:25 +01:00
|
|
|
ret_type, a);
|
|
|
|
}
|
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
/*
|
|
|
|
*/
|
2015-07-21 00:58:43 +01:00
|
|
|
static inline LLVMValueRef
|
2012-11-22 17:37:18 +00:00
|
|
|
lp_build_round_altivec(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
enum lp_build_round_mode mode)
|
|
|
|
{
|
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
const char *intrinsic = NULL;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
|
|
|
|
assert(lp_check_value(type, a));
|
2021-02-25 18:04:50 +00:00
|
|
|
assert(util_get_cpu_caps()->has_altivec);
|
2012-11-22 17:37:18 +00:00
|
|
|
|
2015-03-20 12:02:42 +00:00
|
|
|
(void)type;
|
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
switch (mode) {
|
|
|
|
case LP_BUILD_ROUND_NEAREST:
|
|
|
|
intrinsic = "llvm.ppc.altivec.vrfin";
|
|
|
|
break;
|
|
|
|
case LP_BUILD_ROUND_FLOOR:
|
|
|
|
intrinsic = "llvm.ppc.altivec.vrfim";
|
|
|
|
break;
|
|
|
|
case LP_BUILD_ROUND_CEIL:
|
|
|
|
intrinsic = "llvm.ppc.altivec.vrfip";
|
|
|
|
break;
|
|
|
|
case LP_BUILD_ROUND_TRUNCATE:
|
|
|
|
intrinsic = "llvm.ppc.altivec.vrfiz";
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, bld->vec_type, a);
|
|
|
|
}
|
|
|
|
|
2015-07-21 00:58:43 +01:00
|
|
|
static inline LLVMValueRef
|
2012-11-22 17:37:18 +00:00
|
|
|
lp_build_round_arch(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
enum lp_build_round_mode mode)
|
|
|
|
{
|
2021-11-18 17:27:35 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse4_1 || util_get_cpu_caps()->has_neon ||
|
|
|
|
util_cpu_caps_has_zarch()) {
|
2016-04-01 11:50:28 +01:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
const char *intrinsic_root;
|
|
|
|
char intrinsic[32];
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
(void)type;
|
|
|
|
|
|
|
|
switch (mode) {
|
|
|
|
case LP_BUILD_ROUND_NEAREST:
|
2016-04-13 04:00:03 +01:00
|
|
|
intrinsic_root = "llvm.nearbyint";
|
2016-04-01 11:50:28 +01:00
|
|
|
break;
|
|
|
|
case LP_BUILD_ROUND_FLOOR:
|
|
|
|
intrinsic_root = "llvm.floor";
|
|
|
|
break;
|
|
|
|
case LP_BUILD_ROUND_CEIL:
|
|
|
|
intrinsic_root = "llvm.ceil";
|
|
|
|
break;
|
|
|
|
case LP_BUILD_ROUND_TRUNCATE:
|
|
|
|
intrinsic_root = "llvm.trunc";
|
|
|
|
break;
|
2021-01-26 19:14:02 +00:00
|
|
|
default:
|
|
|
|
unreachable("unhandled lp_build_round_mode");
|
2016-04-01 11:50:28 +01:00
|
|
|
}
|
|
|
|
|
2016-04-03 22:52:53 +01:00
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, intrinsic_root, bld->vec_type);
|
2016-04-01 11:50:28 +01:00
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, bld->vec_type, a);
|
|
|
|
}
|
2021-02-25 18:04:50 +00:00
|
|
|
else /* (util_get_cpu_caps()->has_altivec) */
|
2012-11-22 17:37:18 +00:00
|
|
|
return lp_build_round_altivec(bld, a, mode);
|
|
|
|
}
|
|
|
|
|
2010-07-06 18:33:56 +01:00
|
|
|
/**
|
2010-10-08 17:31:38 +01:00
|
|
|
* Return the integer part of a float (vector) value (== round toward zero).
|
|
|
|
* The returned value is a float (vector).
|
|
|
|
* Ex: trunc(-1.5) = -1.0
|
2010-07-06 18:33:56 +01:00
|
|
|
*/
|
2009-09-29 16:59:13 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_trunc(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-29 16:59:13 +01:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2021-09-07 02:17:39 +01:00
|
|
|
if (type.width == 16) {
|
|
|
|
char intrinsic[64];
|
|
|
|
lp_format_intrinsic(intrinsic, 64, "llvm.trunc", bld->vec_type);
|
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, bld->vec_type, a);
|
|
|
|
}
|
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
|
|
|
return lp_build_round_arch(bld, a, LP_BUILD_ROUND_TRUNCATE);
|
2010-09-29 22:27:38 +01:00
|
|
|
}
|
2009-09-29 16:59:13 +01:00
|
|
|
else {
|
2013-01-31 19:27:49 +00:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
struct lp_type inttype;
|
|
|
|
struct lp_build_context intbld;
|
2014-07-21 16:53:35 +01:00
|
|
|
LLVMValueRef cmpval = lp_build_const_vec(bld->gallivm, type, 1<<24);
|
2013-01-31 19:27:49 +00:00
|
|
|
LLVMValueRef trunc, res, anosign, mask;
|
|
|
|
LLVMTypeRef int_vec_type = bld->int_vec_type;
|
|
|
|
LLVMTypeRef vec_type = bld->vec_type;
|
|
|
|
|
|
|
|
inttype = type;
|
|
|
|
inttype.floating = 0;
|
|
|
|
lp_build_context_init(&intbld, bld->gallivm, inttype);
|
|
|
|
|
|
|
|
/* round by truncation */
|
|
|
|
trunc = LLVMBuildFPToSI(builder, a, int_vec_type, "");
|
|
|
|
res = LLVMBuildSIToFP(builder, trunc, vec_type, "floor.trunc");
|
|
|
|
|
|
|
|
/* mask out sign bit */
|
|
|
|
anosign = lp_build_abs(bld, a);
|
|
|
|
/*
|
|
|
|
* mask out all values if anosign > 2^24
|
|
|
|
* This should work both for large ints (all rounding is no-op for them
|
|
|
|
* because such floats are always exact) as well as special cases like
|
|
|
|
* NaNs, Infs (taking advantage of the fact they use max exponent).
|
|
|
|
* (2^24 is arbitrary anything between 2^24 and 2^31 should work.)
|
|
|
|
*/
|
|
|
|
anosign = LLVMBuildBitCast(builder, anosign, int_vec_type, "");
|
|
|
|
cmpval = LLVMBuildBitCast(builder, cmpval, int_vec_type, "");
|
|
|
|
mask = lp_build_cmp(&intbld, PIPE_FUNC_GREATER, anosign, cmpval);
|
|
|
|
return lp_build_select(bld, mask, a, res);
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-07-06 18:33:56 +01:00
|
|
|
/**
|
|
|
|
* Return float (vector) rounded to nearest integer (vector). The returned
|
|
|
|
* value is a float (vector).
|
|
|
|
* Ex: round(0.9) = 1.0
|
|
|
|
* Ex: round(-1.5) = -2.0
|
|
|
|
*/
|
2009-09-13 13:50:19 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_round(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-09-13 13:50:19 +01:00
|
|
|
|
|
|
|
assert(type.floating);
|
2009-09-29 16:59:13 +01:00
|
|
|
assert(lp_check_value(type, a));
|
2009-09-13 13:50:19 +01:00
|
|
|
|
2021-09-07 02:17:39 +01:00
|
|
|
if (type.width == 16) {
|
|
|
|
char intrinsic[64];
|
|
|
|
lp_format_intrinsic(intrinsic, 64, "llvm.round", bld->vec_type);
|
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, bld->vec_type, a);
|
|
|
|
}
|
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
|
|
|
return lp_build_round_arch(bld, a, LP_BUILD_ROUND_NEAREST);
|
2010-09-29 22:27:38 +01:00
|
|
|
}
|
2009-09-29 16:59:13 +01:00
|
|
|
else {
|
2013-01-31 19:27:49 +00:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
struct lp_type inttype;
|
|
|
|
struct lp_build_context intbld;
|
2014-07-21 16:53:35 +01:00
|
|
|
LLVMValueRef cmpval = lp_build_const_vec(bld->gallivm, type, 1<<24);
|
2013-01-31 19:27:49 +00:00
|
|
|
LLVMValueRef res, anosign, mask;
|
|
|
|
LLVMTypeRef int_vec_type = bld->int_vec_type;
|
|
|
|
LLVMTypeRef vec_type = bld->vec_type;
|
|
|
|
|
|
|
|
inttype = type;
|
|
|
|
inttype.floating = 0;
|
|
|
|
lp_build_context_init(&intbld, bld->gallivm, inttype);
|
|
|
|
|
2009-09-29 16:59:13 +01:00
|
|
|
res = lp_build_iround(bld, a);
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildSIToFP(builder, res, vec_type, "");
|
2013-01-31 19:27:49 +00:00
|
|
|
|
|
|
|
/* mask out sign bit */
|
|
|
|
anosign = lp_build_abs(bld, a);
|
|
|
|
/*
|
|
|
|
* mask out all values if anosign > 2^24
|
|
|
|
* This should work both for large ints (all rounding is no-op for them
|
|
|
|
* because such floats are always exact) as well as special cases like
|
|
|
|
* NaNs, Infs (taking advantage of the fact they use max exponent).
|
|
|
|
* (2^24 is arbitrary anything between 2^24 and 2^31 should work.)
|
|
|
|
*/
|
|
|
|
anosign = LLVMBuildBitCast(builder, anosign, int_vec_type, "");
|
|
|
|
cmpval = LLVMBuildBitCast(builder, cmpval, int_vec_type, "");
|
|
|
|
mask = lp_build_cmp(&intbld, PIPE_FUNC_GREATER, anosign, cmpval);
|
|
|
|
return lp_build_select(bld, mask, a, res);
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
2009-09-13 13:50:19 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-07-06 18:33:56 +01:00
|
|
|
/**
|
|
|
|
* Return floor of float (vector), result is a float (vector)
|
|
|
|
* Ex: floor(1.1) = 1.0
|
|
|
|
* Ex: floor(-1.1) = -2.0
|
|
|
|
*/
|
2009-09-07 14:26:30 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_floor(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-09-07 14:26:30 +01:00
|
|
|
|
|
|
|
assert(type.floating);
|
2010-05-08 23:07:09 +01:00
|
|
|
assert(lp_check_value(type, a));
|
2009-09-07 14:26:30 +01:00
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
|
|
|
return lp_build_round_arch(bld, a, LP_BUILD_ROUND_FLOOR);
|
2010-09-29 22:27:38 +01:00
|
|
|
}
|
2009-09-29 16:59:13 +01:00
|
|
|
else {
|
2013-01-31 19:27:49 +00:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
struct lp_type inttype;
|
|
|
|
struct lp_build_context intbld;
|
2014-07-21 16:53:35 +01:00
|
|
|
LLVMValueRef cmpval = lp_build_const_vec(bld->gallivm, type, 1<<24);
|
2013-01-31 19:27:49 +00:00
|
|
|
LLVMValueRef trunc, res, anosign, mask;
|
|
|
|
LLVMTypeRef int_vec_type = bld->int_vec_type;
|
|
|
|
LLVMTypeRef vec_type = bld->vec_type;
|
|
|
|
|
gallivm: add fp64 support. (v2.1)
This adds support for ARB_gpu_shader_fp64 and ARB_vertex_attrib_64bit to
llvmpipe.
Two things that don't mix well are SoA and doubles, see
emit_fetch_double, and emit_store_double_chan in this.
I've also had to split emit_data.chan, to add src_chan,
which can be different for doubles.
It handles indirect double fetches from temps, inputs, constants
and immediates. It doesn't handle double stores to indirects,
however it appears the mesa/st doesn't currently emit these,
it always does UARL/MOV combos, which will work fine.
tested with piglit, no regressions, all the fp64 tests seem to pass.
v2:
switch to using shuffles for fetch/store (Roland)
assert on indirect double stores - mesa/st never emits these (it uses MOV)
fix indirect temp/input/constant/immediates (Roland)
typos/formatting fixes (Roland)
v2.1:
cleanup some long lines, emit_store_double_chan cleanups.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2015-06-27 05:21:54 +01:00
|
|
|
if (type.width != 32) {
|
|
|
|
char intrinsic[32];
|
2016-04-03 22:52:53 +01:00
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.floor", vec_type);
|
gallivm: add fp64 support. (v2.1)
This adds support for ARB_gpu_shader_fp64 and ARB_vertex_attrib_64bit to
llvmpipe.
Two things that don't mix well are SoA and doubles, see
emit_fetch_double, and emit_store_double_chan in this.
I've also had to split emit_data.chan, to add src_chan,
which can be different for doubles.
It handles indirect double fetches from temps, inputs, constants
and immediates. It doesn't handle double stores to indirects,
however it appears the mesa/st doesn't currently emit these,
it always does UARL/MOV combos, which will work fine.
tested with piglit, no regressions, all the fp64 tests seem to pass.
v2:
switch to using shuffles for fetch/store (Roland)
assert on indirect double stores - mesa/st never emits these (it uses MOV)
fix indirect temp/input/constant/immediates (Roland)
typos/formatting fixes (Roland)
v2.1:
cleanup some long lines, emit_store_double_chan cleanups.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2015-06-27 05:21:54 +01:00
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a);
|
|
|
|
}
|
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
assert(type.width == 32); /* might want to handle doubles at some point */
|
|
|
|
|
|
|
|
inttype = type;
|
|
|
|
inttype.floating = 0;
|
|
|
|
lp_build_context_init(&intbld, bld->gallivm, inttype);
|
|
|
|
|
|
|
|
/* round by truncation */
|
|
|
|
trunc = LLVMBuildFPToSI(builder, a, int_vec_type, "");
|
|
|
|
res = LLVMBuildSIToFP(builder, trunc, vec_type, "floor.trunc");
|
|
|
|
|
|
|
|
if (type.sign) {
|
|
|
|
LLVMValueRef tmp;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* fix values if rounding is wrong (for non-special cases)
|
|
|
|
* - this is the case if trunc > a
|
|
|
|
*/
|
|
|
|
mask = lp_build_cmp(bld, PIPE_FUNC_GREATER, res, a);
|
|
|
|
/* tmp = trunc > a ? 1.0 : 0.0 */
|
|
|
|
tmp = LLVMBuildBitCast(builder, bld->one, int_vec_type, "");
|
|
|
|
tmp = lp_build_and(&intbld, mask, tmp);
|
|
|
|
tmp = LLVMBuildBitCast(builder, tmp, vec_type, "");
|
|
|
|
res = lp_build_sub(bld, res, tmp);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* mask out sign bit */
|
|
|
|
anosign = lp_build_abs(bld, a);
|
|
|
|
/*
|
|
|
|
* mask out all values if anosign > 2^24
|
|
|
|
* This should work both for large ints (all rounding is no-op for them
|
|
|
|
* because such floats are always exact) as well as special cases like
|
|
|
|
* NaNs, Infs (taking advantage of the fact they use max exponent).
|
|
|
|
* (2^24 is arbitrary anything between 2^24 and 2^31 should work.)
|
|
|
|
*/
|
|
|
|
anosign = LLVMBuildBitCast(builder, anosign, int_vec_type, "");
|
|
|
|
cmpval = LLVMBuildBitCast(builder, cmpval, int_vec_type, "");
|
|
|
|
mask = lp_build_cmp(&intbld, PIPE_FUNC_GREATER, anosign, cmpval);
|
|
|
|
return lp_build_select(bld, mask, a, res);
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
2009-09-07 14:26:30 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-07-06 18:33:56 +01:00
|
|
|
/**
|
|
|
|
* Return ceiling of float (vector), returning float (vector).
|
|
|
|
* Ex: ceil( 1.1) = 2.0
|
|
|
|
* Ex: ceil(-1.1) = -1.0
|
|
|
|
*/
|
2009-09-13 13:50:19 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_ceil(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-09-13 13:50:19 +01:00
|
|
|
|
|
|
|
assert(type.floating);
|
2009-09-29 16:59:13 +01:00
|
|
|
assert(lp_check_value(type, a));
|
2009-09-13 13:50:19 +01:00
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
|
|
|
return lp_build_round_arch(bld, a, LP_BUILD_ROUND_CEIL);
|
2010-09-29 22:27:38 +01:00
|
|
|
}
|
2009-09-29 16:59:13 +01:00
|
|
|
else {
|
2013-01-31 19:27:49 +00:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
struct lp_type inttype;
|
|
|
|
struct lp_build_context intbld;
|
2014-07-21 16:53:35 +01:00
|
|
|
LLVMValueRef cmpval = lp_build_const_vec(bld->gallivm, type, 1<<24);
|
2013-01-31 19:27:49 +00:00
|
|
|
LLVMValueRef trunc, res, anosign, mask, tmp;
|
|
|
|
LLVMTypeRef int_vec_type = bld->int_vec_type;
|
|
|
|
LLVMTypeRef vec_type = bld->vec_type;
|
|
|
|
|
gallivm: add fp64 support. (v2.1)
This adds support for ARB_gpu_shader_fp64 and ARB_vertex_attrib_64bit to
llvmpipe.
Two things that don't mix well are SoA and doubles, see
emit_fetch_double, and emit_store_double_chan in this.
I've also had to split emit_data.chan, to add src_chan,
which can be different for doubles.
It handles indirect double fetches from temps, inputs, constants
and immediates. It doesn't handle double stores to indirects,
however it appears the mesa/st doesn't currently emit these,
it always does UARL/MOV combos, which will work fine.
tested with piglit, no regressions, all the fp64 tests seem to pass.
v2:
switch to using shuffles for fetch/store (Roland)
assert on indirect double stores - mesa/st never emits these (it uses MOV)
fix indirect temp/input/constant/immediates (Roland)
typos/formatting fixes (Roland)
v2.1:
cleanup some long lines, emit_store_double_chan cleanups.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2015-06-27 05:21:54 +01:00
|
|
|
if (type.width != 32) {
|
|
|
|
char intrinsic[32];
|
2016-04-03 22:52:53 +01:00
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.ceil", vec_type);
|
gallivm: add fp64 support. (v2.1)
This adds support for ARB_gpu_shader_fp64 and ARB_vertex_attrib_64bit to
llvmpipe.
Two things that don't mix well are SoA and doubles, see
emit_fetch_double, and emit_store_double_chan in this.
I've also had to split emit_data.chan, to add src_chan,
which can be different for doubles.
It handles indirect double fetches from temps, inputs, constants
and immediates. It doesn't handle double stores to indirects,
however it appears the mesa/st doesn't currently emit these,
it always does UARL/MOV combos, which will work fine.
tested with piglit, no regressions, all the fp64 tests seem to pass.
v2:
switch to using shuffles for fetch/store (Roland)
assert on indirect double stores - mesa/st never emits these (it uses MOV)
fix indirect temp/input/constant/immediates (Roland)
typos/formatting fixes (Roland)
v2.1:
cleanup some long lines, emit_store_double_chan cleanups.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2015-06-27 05:21:54 +01:00
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a);
|
|
|
|
}
|
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
assert(type.width == 32); /* might want to handle doubles at some point */
|
|
|
|
|
|
|
|
inttype = type;
|
|
|
|
inttype.floating = 0;
|
|
|
|
lp_build_context_init(&intbld, bld->gallivm, inttype);
|
|
|
|
|
|
|
|
/* round by truncation */
|
|
|
|
trunc = LLVMBuildFPToSI(builder, a, int_vec_type, "");
|
|
|
|
trunc = LLVMBuildSIToFP(builder, trunc, vec_type, "ceil.trunc");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* fix values if rounding is wrong (for non-special cases)
|
|
|
|
* - this is the case if trunc < a
|
|
|
|
*/
|
|
|
|
mask = lp_build_cmp(bld, PIPE_FUNC_LESS, trunc, a);
|
|
|
|
/* tmp = trunc < a ? 1.0 : 0.0 */
|
|
|
|
tmp = LLVMBuildBitCast(builder, bld->one, int_vec_type, "");
|
|
|
|
tmp = lp_build_and(&intbld, mask, tmp);
|
|
|
|
tmp = LLVMBuildBitCast(builder, tmp, vec_type, "");
|
|
|
|
res = lp_build_add(bld, trunc, tmp);
|
|
|
|
|
|
|
|
/* mask out sign bit */
|
|
|
|
anosign = lp_build_abs(bld, a);
|
|
|
|
/*
|
|
|
|
* mask out all values if anosign > 2^24
|
|
|
|
* This should work both for large ints (all rounding is no-op for them
|
|
|
|
* because such floats are always exact) as well as special cases like
|
|
|
|
* NaNs, Infs (taking advantage of the fact they use max exponent).
|
|
|
|
* (2^24 is arbitrary anything between 2^24 and 2^31 should work.)
|
|
|
|
*/
|
|
|
|
anosign = LLVMBuildBitCast(builder, anosign, int_vec_type, "");
|
|
|
|
cmpval = LLVMBuildBitCast(builder, cmpval, int_vec_type, "");
|
|
|
|
mask = lp_build_cmp(&intbld, PIPE_FUNC_GREATER, anosign, cmpval);
|
|
|
|
return lp_build_select(bld, mask, a, res);
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
2009-09-13 13:50:19 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-03-04 17:50:26 +00:00
|
|
|
/**
|
2010-07-06 18:33:56 +01:00
|
|
|
* Return fractional part of 'a' computed as a - floor(a)
|
2010-03-04 17:50:26 +00:00
|
|
|
* Typically used in texture coord arithmetic.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_fract(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
|
|
|
assert(bld->type.floating);
|
|
|
|
return lp_build_sub(bld, a, lp_build_floor(bld, a));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
/**
|
2016-04-26 03:53:01 +01:00
|
|
|
* Prevent returning 1.0 for very small negative values of 'a' by clamping
|
|
|
|
* against 0.99999(9). (Will also return that value for NaNs.)
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
*/
|
|
|
|
static inline LLVMValueRef
|
|
|
|
clamp_fract(struct lp_build_context *bld, LLVMValueRef fract)
|
|
|
|
{
|
|
|
|
LLVMValueRef max;
|
|
|
|
|
|
|
|
/* this is the largest number smaller than 1.0 representable as float */
|
|
|
|
max = lp_build_const_vec(bld->gallivm, bld->type,
|
|
|
|
1.0 - 1.0/(1LL << (lp_mantissa(bld->type) + 1)));
|
2016-04-26 03:53:01 +01:00
|
|
|
return lp_build_min_ext(bld, fract, max,
|
|
|
|
GALLIVM_NAN_RETURN_OTHER_SECOND_NONNAN);
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Same as lp_build_fract, but guarantees that the result is always smaller
|
2016-04-26 03:53:01 +01:00
|
|
|
* than one. Will also return the smaller-than-one value for infs, NaNs.
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_fract_safe(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
|
|
|
return clamp_fract(bld, lp_build_fract(bld, a));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-09-29 16:59:13 +01:00
|
|
|
/**
|
2010-10-08 17:31:38 +01:00
|
|
|
* Return the integer part of a float (vector) value (== round toward zero).
|
|
|
|
* The returned value is an integer (vector).
|
|
|
|
* Ex: itrunc(-1.5) = -1
|
2009-09-29 16:59:13 +01:00
|
|
|
*/
|
2009-09-13 13:50:19 +01:00
|
|
|
LLVMValueRef
|
2009-09-29 16:59:13 +01:00
|
|
|
lp_build_itrunc(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
2009-09-13 13:50:19 +01:00
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
|
2009-09-13 13:50:19 +01:00
|
|
|
|
|
|
|
assert(type.floating);
|
2010-05-08 23:07:09 +01:00
|
|
|
assert(lp_check_value(type, a));
|
2009-09-13 13:50:19 +01:00
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildFPToSI(builder, a, int_vec_type, "");
|
2009-09-13 13:50:19 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-03-11 01:07:13 +00:00
|
|
|
/**
|
2010-07-06 18:33:56 +01:00
|
|
|
* Return float (vector) rounded to nearest integer (vector). The returned
|
|
|
|
* value is an integer (vector).
|
|
|
|
* Ex: iround(0.9) = 1
|
|
|
|
* Ex: iround(-1.5) = -2
|
2010-03-11 01:07:13 +00:00
|
|
|
*/
|
2009-09-07 14:26:30 +01:00
|
|
|
LLVMValueRef
|
2009-09-29 16:59:13 +01:00
|
|
|
lp_build_iround(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
2009-09-07 14:26:30 +01:00
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-10-06 12:09:32 +01:00
|
|
|
LLVMTypeRef int_vec_type = bld->int_vec_type;
|
2009-09-29 16:59:13 +01:00
|
|
|
LLVMValueRef res;
|
2009-09-07 14:26:30 +01:00
|
|
|
|
|
|
|
assert(type.floating);
|
2010-03-09 16:37:50 +00:00
|
|
|
|
2009-09-29 16:59:13 +01:00
|
|
|
assert(lp_check_value(type, a));
|
2009-09-07 14:26:30 +01:00
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if ((util_get_cpu_caps()->has_sse2 &&
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
((type.width == 32) && (type.length == 1 || type.length == 4))) ||
|
2021-02-25 18:04:50 +00:00
|
|
|
(util_get_cpu_caps()->has_avx && type.width == 32 && type.length == 8)) {
|
2010-10-08 17:38:25 +01:00
|
|
|
return lp_build_iround_nearest_sse2(bld, a);
|
|
|
|
}
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
|
|
|
res = lp_build_round_arch(bld, a, LP_BUILD_ROUND_NEAREST);
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
|
|
|
else {
|
|
|
|
LLVMValueRef half;
|
|
|
|
|
2018-11-21 20:13:19 +00:00
|
|
|
half = lp_build_const_vec(bld->gallivm, type, nextafterf(0.5, 0.0));
|
2010-10-06 12:09:32 +01:00
|
|
|
|
|
|
|
if (type.sign) {
|
|
|
|
LLVMTypeRef vec_type = bld->vec_type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMValueRef mask = lp_build_const_int_vec(bld->gallivm, type,
|
|
|
|
(unsigned long long)1 << (type.width - 1));
|
2010-10-06 12:09:32 +01:00
|
|
|
LLVMValueRef sign;
|
|
|
|
|
|
|
|
/* get sign bit */
|
2010-12-03 01:11:13 +00:00
|
|
|
sign = LLVMBuildBitCast(builder, a, int_vec_type, "");
|
|
|
|
sign = LLVMBuildAnd(builder, sign, mask, "");
|
2010-10-06 12:09:32 +01:00
|
|
|
|
|
|
|
/* sign * 0.5 */
|
2010-12-03 01:11:13 +00:00
|
|
|
half = LLVMBuildBitCast(builder, half, int_vec_type, "");
|
|
|
|
half = LLVMBuildOr(builder, sign, half, "");
|
|
|
|
half = LLVMBuildBitCast(builder, half, vec_type, "");
|
2010-10-06 12:09:32 +01:00
|
|
|
}
|
2009-09-29 16:59:13 +01:00
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFAdd(builder, a, half, "");
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
|
2009-09-29 16:59:13 +01:00
|
|
|
|
|
|
|
return res;
|
2009-09-07 14:26:30 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-01-13 16:31:36 +00:00
|
|
|
/**
|
2010-07-06 18:33:56 +01:00
|
|
|
* Return floor of float (vector), result is an int (vector)
|
|
|
|
* Ex: ifloor(1.1) = 1.0
|
|
|
|
* Ex: ifloor(-1.1) = -2.0
|
2010-01-13 16:31:36 +00:00
|
|
|
*/
|
2009-09-07 14:26:30 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_ifloor(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-29 16:59:13 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-10-06 12:09:32 +01:00
|
|
|
LLVMTypeRef int_vec_type = bld->int_vec_type;
|
2009-09-29 16:59:13 +01:00
|
|
|
LLVMValueRef res;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
res = a;
|
|
|
|
if (type.sign) {
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
|
|
|
res = lp_build_round_arch(bld, a, LP_BUILD_ROUND_FLOOR);
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
}
|
|
|
|
else {
|
2013-01-31 19:27:49 +00:00
|
|
|
struct lp_type inttype;
|
|
|
|
struct lp_build_context intbld;
|
|
|
|
LLVMValueRef trunc, itrunc, mask;
|
2010-10-06 12:09:32 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
assert(type.floating);
|
|
|
|
assert(lp_check_value(type, a));
|
2010-10-06 12:09:32 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
inttype = type;
|
|
|
|
inttype.floating = 0;
|
|
|
|
lp_build_context_init(&intbld, bld->gallivm, inttype);
|
2010-10-06 12:09:32 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
/* round by truncation */
|
|
|
|
itrunc = LLVMBuildFPToSI(builder, a, int_vec_type, "");
|
|
|
|
trunc = LLVMBuildSIToFP(builder, itrunc, bld->vec_type, "ifloor.trunc");
|
2010-10-06 12:09:32 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
/*
|
|
|
|
* fix values if rounding is wrong (for non-special cases)
|
|
|
|
* - this is the case if trunc > a
|
|
|
|
* The results of doing this with NaNs, very large values etc.
|
|
|
|
* are undefined but this seems to be the case anyway.
|
|
|
|
*/
|
|
|
|
mask = lp_build_cmp(bld, PIPE_FUNC_GREATER, trunc, a);
|
|
|
|
/* cheapie minus one with mask since the mask is minus one / zero */
|
|
|
|
return lp_build_add(&intbld, itrunc, mask);
|
2010-10-06 12:09:32 +01:00
|
|
|
}
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
|
|
|
|
2010-07-06 18:33:56 +01:00
|
|
|
/* round to nearest (toward zero) */
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFPToSI(builder, res, int_vec_type, "ifloor.res");
|
2009-09-29 16:59:13 +01:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-07-06 18:33:56 +01:00
|
|
|
/**
|
|
|
|
* Return ceiling of float (vector), returning int (vector).
|
|
|
|
* Ex: iceil( 1.1) = 2
|
|
|
|
* Ex: iceil(-1.1) = -1
|
|
|
|
*/
|
2009-09-29 16:59:13 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_iceil(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-29 16:59:13 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-10-06 12:09:32 +01:00
|
|
|
LLVMTypeRef int_vec_type = bld->int_vec_type;
|
2009-09-29 16:59:13 +01:00
|
|
|
LLVMValueRef res;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
|
|
|
res = lp_build_round_arch(bld, a, LP_BUILD_ROUND_CEIL);
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
|
|
|
else {
|
2013-01-31 19:27:49 +00:00
|
|
|
struct lp_type inttype;
|
|
|
|
struct lp_build_context intbld;
|
|
|
|
LLVMValueRef trunc, itrunc, mask;
|
2010-07-06 18:33:56 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
assert(type.floating);
|
|
|
|
assert(lp_check_value(type, a));
|
2010-07-06 18:33:56 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
inttype = type;
|
|
|
|
inttype.floating = 0;
|
|
|
|
lp_build_context_init(&intbld, bld->gallivm, inttype);
|
2010-10-06 12:09:32 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
/* round by truncation */
|
|
|
|
itrunc = LLVMBuildFPToSI(builder, a, int_vec_type, "");
|
|
|
|
trunc = LLVMBuildSIToFP(builder, itrunc, bld->vec_type, "iceil.trunc");
|
2010-07-06 18:33:56 +01:00
|
|
|
|
2013-01-31 19:27:49 +00:00
|
|
|
/*
|
|
|
|
* fix values if rounding is wrong (for non-special cases)
|
|
|
|
* - this is the case if trunc < a
|
|
|
|
* The results of doing this with NaNs, very large values etc.
|
|
|
|
* are undefined but this seems to be the case anyway.
|
|
|
|
*/
|
|
|
|
mask = lp_build_cmp(bld, PIPE_FUNC_LESS, trunc, a);
|
|
|
|
/* cheapie plus one with mask since the mask is minus one / zero */
|
|
|
|
return lp_build_sub(&intbld, itrunc, mask);
|
2009-09-29 16:59:13 +01:00
|
|
|
}
|
|
|
|
|
2010-07-06 18:33:56 +01:00
|
|
|
/* round to nearest (toward zero) */
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFPToSI(builder, res, int_vec_type, "iceil.res");
|
2009-09-29 16:59:13 +01:00
|
|
|
|
|
|
|
return res;
|
2009-09-07 14:26:30 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-10-06 18:31:36 +01:00
|
|
|
/**
|
|
|
|
* Combined ifloor() & fract().
|
|
|
|
*
|
|
|
|
* Preferred to calling the functions separately, as it will ensure that the
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
* strategy (floor() vs ifloor()) that results in less redundant work is used.
|
2010-10-06 18:31:36 +01:00
|
|
|
*/
|
|
|
|
void
|
|
|
|
lp_build_ifloor_fract(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef *out_ipart,
|
|
|
|
LLVMValueRef *out_fpart)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-10-06 18:31:36 +01:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
LLVMValueRef ipart;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2012-11-22 17:37:18 +00:00
|
|
|
if (arch_rounding_available(type)) {
|
2010-10-06 18:31:36 +01:00
|
|
|
/*
|
|
|
|
* floor() is easier.
|
|
|
|
*/
|
|
|
|
|
|
|
|
ipart = lp_build_floor(bld, a);
|
2010-12-03 01:11:13 +00:00
|
|
|
*out_fpart = LLVMBuildFSub(builder, a, ipart, "fpart");
|
|
|
|
*out_ipart = LLVMBuildFPToSI(builder, ipart, bld->int_vec_type, "ipart");
|
2010-10-06 18:31:36 +01:00
|
|
|
}
|
|
|
|
else {
|
|
|
|
/*
|
|
|
|
* ifloor() is easier.
|
|
|
|
*/
|
|
|
|
|
|
|
|
*out_ipart = lp_build_ifloor(bld, a);
|
2010-12-03 01:11:13 +00:00
|
|
|
ipart = LLVMBuildSIToFP(builder, *out_ipart, bld->vec_type, "ipart");
|
|
|
|
*out_fpart = LLVMBuildFSub(builder, a, ipart, "fpart");
|
2010-10-06 18:31:36 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
/**
|
|
|
|
* Same as lp_build_ifloor_fract, but guarantees that the fractional part is
|
|
|
|
* always smaller than one.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
lp_build_ifloor_fract_safe(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef *out_ipart,
|
|
|
|
LLVMValueRef *out_fpart)
|
|
|
|
{
|
|
|
|
lp_build_ifloor_fract(bld, a, out_ipart, out_fpart);
|
|
|
|
*out_fpart = clamp_fract(bld, *out_fpart);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_sqrt(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
2009-08-12 02:34:55 +01:00
|
|
|
char intrinsic[32];
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
assert(type.floating);
|
2016-04-03 22:52:53 +01:00
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.sqrt", vec_type);
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, vec_type, a);
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-08-14 18:02:47 +01:00
|
|
|
/**
|
|
|
|
* Do one Newton-Raphson step to improve reciprocate precision:
|
|
|
|
*
|
2019-05-31 17:10:40 +01:00
|
|
|
* x_{i+1} = x_i + x_i * (1 - a * x_i)
|
2010-08-14 18:02:47 +01:00
|
|
|
*
|
2010-08-21 21:58:22 +01:00
|
|
|
* XXX: Unfortunately this won't give IEEE-754 conformant results for 0 or
|
|
|
|
* +/-Inf, giving NaN instead. Certain applications rely on this behavior,
|
2019-05-31 17:10:40 +01:00
|
|
|
* such as Google Earth, which does RCP(RSQRT(0.0)) when drawing the Earth's
|
2010-08-21 21:58:22 +01:00
|
|
|
* halo. It would be necessary to clamp the argument to prevent this.
|
|
|
|
*
|
2010-08-14 18:02:47 +01:00
|
|
|
* See also:
|
|
|
|
* - http://en.wikipedia.org/wiki/Division_(digital)#Newton.E2.80.93Raphson_division
|
|
|
|
* - http://softwarecommunity.intel.com/articles/eng/1818.htm
|
|
|
|
*/
|
2015-07-21 00:58:43 +01:00
|
|
|
static inline LLVMValueRef
|
2010-08-14 18:02:47 +01:00
|
|
|
lp_build_rcp_refine(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef rcp_a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2019-05-31 17:10:40 +01:00
|
|
|
LLVMValueRef neg_a;
|
2010-08-14 18:02:47 +01:00
|
|
|
LLVMValueRef res;
|
|
|
|
|
2019-05-31 17:10:40 +01:00
|
|
|
neg_a = LLVMBuildFNeg(builder, a, "");
|
|
|
|
res = lp_build_fmuladd(builder, neg_a, rcp_a, bld->one);
|
|
|
|
res = lp_build_fmuladd(builder, res, rcp_a, rcp_a);
|
2010-08-14 18:02:47 +01:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_rcp(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
if(a == bld->zero)
|
|
|
|
return bld->undef;
|
|
|
|
if(a == bld->one)
|
|
|
|
return bld->one;
|
|
|
|
if(a == bld->undef)
|
|
|
|
return bld->undef;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
|
|
|
|
if(LLVMIsConstant(a))
|
|
|
|
return LLVMConstFDiv(bld->one, a);
|
|
|
|
|
2010-08-21 21:58:22 +01:00
|
|
|
/*
|
|
|
|
* We don't use RCPPS because:
|
|
|
|
* - it only has 10bits of precision
|
|
|
|
* - it doesn't even get the reciprocate of 1.0 exactly
|
|
|
|
* - doing Newton-Rapshon steps yields wrong (NaN) values for 0.0 or Inf
|
|
|
|
* - for recent processors the benefit over DIVPS is marginal, a case
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
* dependent
|
2010-08-21 21:58:22 +01:00
|
|
|
*
|
|
|
|
* We could still use it on certain processors if benchmarks show that the
|
|
|
|
* RCPPS plus necessary workarounds are still preferrable to DIVPS; or for
|
|
|
|
* particular uses that require less workarounds.
|
|
|
|
*/
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if (FALSE && ((util_get_cpu_caps()->has_sse && type.width == 32 && type.length == 4) ||
|
|
|
|
(util_get_cpu_caps()->has_avx && type.width == 32 && type.length == 8))){
|
2010-08-21 21:58:22 +01:00
|
|
|
const unsigned num_iterations = 0;
|
2010-05-03 17:06:57 +01:00
|
|
|
LLVMValueRef res;
|
2010-08-14 18:02:47 +01:00
|
|
|
unsigned i;
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
const char *intrinsic = NULL;
|
|
|
|
|
|
|
|
if (type.length == 4) {
|
|
|
|
intrinsic = "llvm.x86.sse.rcp.ps";
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
intrinsic = "llvm.x86.avx.rcp.ps.256";
|
|
|
|
}
|
2010-05-03 17:06:57 +01:00
|
|
|
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
res = lp_build_intrinsic_unary(builder, intrinsic, bld->vec_type, a);
|
2010-05-03 17:06:57 +01:00
|
|
|
|
2010-08-21 21:58:22 +01:00
|
|
|
for (i = 0; i < num_iterations; ++i) {
|
2010-08-14 18:02:47 +01:00
|
|
|
res = lp_build_rcp_refine(bld, a, res);
|
|
|
|
}
|
2010-05-03 17:06:57 +01:00
|
|
|
|
2010-08-11 18:44:17 +01:00
|
|
|
return res;
|
2010-05-03 17:06:57 +01:00
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildFDiv(builder, bld->one, a, "");
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-08-14 18:02:47 +01:00
|
|
|
/**
|
|
|
|
* Do one Newton-Raphson step to improve rsqrt precision:
|
|
|
|
*
|
|
|
|
* x_{i+1} = 0.5 * x_i * (3.0 - a * x_i * x_i)
|
|
|
|
*
|
2012-09-21 16:03:48 +01:00
|
|
|
* See also Intel 64 and IA-32 Architectures Optimization Manual.
|
2010-08-14 18:02:47 +01:00
|
|
|
*/
|
2015-07-21 00:58:43 +01:00
|
|
|
static inline LLVMValueRef
|
2010-08-14 18:02:47 +01:00
|
|
|
lp_build_rsqrt_refine(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
LLVMValueRef rsqrt_a)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMValueRef half = lp_build_const_vec(bld->gallivm, bld->type, 0.5);
|
|
|
|
LLVMValueRef three = lp_build_const_vec(bld->gallivm, bld->type, 3.0);
|
2010-08-14 18:02:47 +01:00
|
|
|
LLVMValueRef res;
|
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildFMul(builder, rsqrt_a, rsqrt_a, "");
|
|
|
|
res = LLVMBuildFMul(builder, a, res, "");
|
|
|
|
res = LLVMBuildFSub(builder, three, res, "");
|
|
|
|
res = LLVMBuildFMul(builder, rsqrt_a, res, "");
|
|
|
|
res = LLVMBuildFMul(builder, half, res, "");
|
2010-08-14 18:02:47 +01:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
2012-09-21 16:03:48 +01:00
|
|
|
* Generate 1/sqrt(a).
|
|
|
|
* Result is undefined for values < 0, infinity for +0.
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2009-08-11 13:09:35 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_rsqrt(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
assert(type.floating);
|
|
|
|
|
2012-09-21 16:03:48 +01:00
|
|
|
/*
|
|
|
|
* This should be faster but all denormals will end up as infinity.
|
|
|
|
*/
|
2013-07-11 22:15:44 +01:00
|
|
|
if (0 && lp_build_fast_rsqrt_available(type)) {
|
2011-07-20 22:41:17 +01:00
|
|
|
const unsigned num_iterations = 1;
|
2010-08-14 18:02:47 +01:00
|
|
|
LLVMValueRef res;
|
|
|
|
unsigned i;
|
gallivm,draw,llvmpipe: Support wider native registers.
Squashed commit of the following:
commit 7acb7b4f60dc505af3dd00dcff744f80315d5b0e
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:46:31 2012 +0100
draw: Don't use dynamically sized arrays.
Not supported by MSVC.
commit 5810c28c83647612cb372d1e763fd9d7780df3cb
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:44:16 2012 +0100
gallivm,llvmpipe: Don't use expressions with PIPE_ALIGN_VAR().
MSVC doesn't accept exceptions in _declspec(align(...)). Use a
define instead.
commit 8aafd1457ba572a02b289b3f3411e99a3c056072
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 9 17:41:56 2012 +0100
gallium/util: Make u_cpu_detect.h header C++ safe.
commit 5795248350771f899cfbfc1a3a58f1835eb2671d
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Jul 2 12:08:01 2012 +0100
gallium/util: Add ULL suffix to large constants.
As suggested by Andy Furniss: it looks like some old gcc versions
require it.
commit 4c66c22727eff92226544c7d43c4eb94de359e10
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Truly disable INF/NAN tests on MSVC.
Thanks to Brian for spotting this.
commit 8bce274c7fad578d7eb656d9a1413f5c0844c94e
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 13:39:07 2012 +0100
gallium/util: Disable INF/NAN tests on MSVC.
Somehow they are not recognized as constants.
commit 6868649cff8d7fd2e2579c28d0b74ef6dd4f9716
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 15:05:24 2012 +0200
gallivm: Cleanup the 2 x 8 float -> 16 ub special path in lp_build_conv.
No behaviour change intended, like 7b98455fb40c2df84cfd3cdb1eb7650f67c8a751.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 5147a0949c4407e8bce9e41d9859314b4a9ccf77
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jul 5 14:28:19 2012 +0200
gallivm: (trivial) fix issues with multiple-of-4 texture fetch
Some formats can't handle non-multiple of 4 fetches I believe, but
everything must support length 1 and multiples of 4.
So avoid going to scalar fetch (which is very costly) just because length
isn't 4.
Also extend the hack to not use shift with variable count for yuv formats to
arbitrary length (larger than 1) - doesn't matter how many elements we
have we always want to avoid it unless we have variable shift count
instruction (which we should get with avx2).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 87ebcb1bd71fa4c739451ec8ca89a7f29b168c08
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jul 4 02:09:55 2012 +0200
gallivm: (trivial) fix typo for wrap repeat mode in linear filtering aos code
This would lead to bogus coordinates at the edges.
(undetected by piglit because this path is only taken for block-based
formats).
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 3a42717101b1619874c8932a580c0b9e6896b557
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue Jul 3 19:42:49 2012 +0100
gallivm: Fix TGSI integer translation with AVX.
commit d71ff104085c196b16426081098fb0bde128ce4f
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Jun 29 15:17:41 2012 +0100
llvmpipe: Fix LLVM JIT linear path.
It was not working properly because it was looking at the JIT function
before it was actually compiled.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit a94df0386213e1f5f9a6ed470c535f9688ec0a1b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Jun 28 18:07:10 2012 +0100
gallivm: Refactor lp_build_broadcast(_scalar) to share code.
Doesn't really change the generated assembly, but produces more compact IR,
and of course, makes code more consistent.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 66712ba2731fc029fa246d4fc477d61ab785edb5
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 17:30:13 2012 +0100
gallivm: Make LLVMContextRef a singleton.
There are any places inside LLVM that depend on it. Too many to attempt
to fix.
Reviewed-by: Brian Paul <brianp@vmware.com>
commit ff5fb7897495ac263f0b069370fab701b70dccef
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 28 18:15:27 2012 +0200
gallivm: don't use 8-wide texture fetch in aos path
This appears to be a slight loss usually.
There are probably several reasons for that:
- fetching itself is scalar
- filtering is pure int code hence needs splitting anyway, same
for the final texel offset calculations
- texture wrap related code, which can be done 8-wide, is slightly more
complex with floats (with clamp_to_edge) and float operations generally
more costly hence probably not much faster overall
- the code needed to split when encountering different mip levels for the
quads, adding complexity
So, just split always for aos path (but leave it 8-wide for soa, since we
do 8-wide filtering there when possible).
This should certainly be revisited if we'd have avx2 support.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ce8032b43dcd8e8d816cbab6428f54b0798f945d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:41:19 2012 +0200
gallivm: (trivial) don't extract fparts variable if not needed
Did not have any consequences but unnecessary.
commit aaa9aaed8f80dc282492f62aa583a7ee23a4c6d5
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 18:09:06 2012 +0200
gallivm: fix precision issue in aos linear int wrap code
now not just passes at a quick glance but also with piglit...
If we do the wrapping with floats, we also need to set the
weights accordingly. We can potentially end up with different
(integer) coordinates than what the integer calculations would
have chosen, which means the integer weights calculated previously
in this case are completely wrong. Well at least that's what I think
happens, at least recalculating the weights helps.
(Some day really should refactor all the wrapping, so we do whatever is
fastest independent of 16bit int aos or 32bit float soa filtering.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit fd6f18588ced7ac8e081892f3bab2916623ad7a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 27 11:15:53 2012 +0100
gallium/util: Fix parsing of options with underscore.
For example
GALLIVM_DEBUG=no_brilinear
which was being parsed as two options, "no" and "brilinear".
commit 09a8f809088178a03e49e409fa18f1ac89561837
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 26 15:00:14 2012 +0100
gallivm: Added a generic lp_build_print_value which prints a LLVMValueRef.
Updated lp_build_printf to share common code.
Removed specific lp_build_print_vecX.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit e59bdcc2c075931bfba2a84967a5ecd1dedd6eb0
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed May 16 15:00:23 2012 +0100
draw,llvmpipe: Avoid named struct types on LLVM 3.0 and later.
Starting with LLVM 3.0, named structures are meant not for debugging, but
for recursive data types, previously also known as opaque types.
The recursive nature of these types leads to several memory management
difficulties. Given that we don't actually need recursive types, avoid
them altogether.
This is an attempt to address fdo bugs 41791 and 44466. The issue is
somewhat random so there's no easy way to check how effective this is.
Cherry-picked from 9af1ba565dfd5cef9ee938bb7c04767d14878fbf
commit df6070f618a203c7a876d984c847cde4cbc26bdb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 27 14:42:53 2012 +0200
gallivm: (trivial) fix typo in faster aos linear int wrap code
no longer crashes, now REALLY tested.
commit d8f98dce452c867214e6782e86dc08562643c862
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 18:20:58 2012 +0200
llvmpipe: (trivial) remove bogus optimization for float aos repeat wrap
This optimization for nearest filtering on the linear path generated
likely bogus results, and the int path didn't have any optimizations
there since the only shader using force_nearest apparently uses
clamp_to_edge not repeat wrap anyway.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit c4e271a0631087c795e756a5bb6b046043b5099d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 23:01:52 2012 +0200
gallivm: faster repeat wrap for linear aos path too
Even if we already have scaled integer coords, it's way faster to use
the original float coord (plus some conversions) rather than use URem.
The choice of what to do for texture wrapping is not really tied to int
aos or float soa filtering though for some modes there can be some gains
(because of easier weight calculations).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 1174a75b1806e92aee4264ffe0ffe7e70abbbfa3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 14:39:22 2012 +0200
gallivm: improve npot tex wrap repeat in linear soa path
URem gets translated into series of scalar divisions so
just about anything else is faster.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit f849ffaa499ed96fa0efd3594fce255c7f22891b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 26 00:40:35 2012 +0100
gallivm: (trivial) fix near-invisible shift-space typo
I blame the keyboard.
commit 5298a0b19fe672aebeb70964c0797d5921b51cf0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:24:28 2012 +0200
gallivm: add new intrinsic helper to deal with arbitrary vector length
This helper will split vectors which are too large for the hw, or expand
them if they are too small, so a caller of a function using intrinsics which
uses such sizes need not split (or expand) the vectors manually and the
function will still use the intrinsic instead of dropping back to generic
llvm code. It can also accept scalars for use with pseudo-vector intrinsics
(only useful for float arguments, all x86 scalar simd float intrinsics use
4vf32).
Only used for lp_build_min/max() for now (also added the scalar float case
for these while there). (Other basic binary functions could use it easily,
whereas functions with a different interface would need different helpers.)
Expanding vectors isn't widely used, because we always try to use
build contexts with native hw vector sizes. But it might (or not) be nicer
if this wouldn't need to be done, the generated code should in theory stay
the same (it does get hit by lp_build_rho though already since we
didn't have a intrinsic for the scalar lp_build_max case before).
v2: incorporated Brian's feedback, and also made the scalar min/max case work
instead of crash (all scalar simd float intrinsics take 4vf32 as argument,
probably the reason why it wasn't used before).
Moved to lp_bld_intr based on José's request, and passing intrinsic size
instead of length.
Ideally we'd derive the source type info from the passed in llvm value refs
and process some llvmtype return type so we could handle intrinsics where
the source and destination type isn't the same (like float/int conversions,
packing instructions) but that's a bit too complicated for now.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 01aa760b99ec0b2dc8ce57a43650e83f8c1becdf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 16:19:18 2012 +0200
gallivm: (trivial) increase max code size for shader disassembly
64kB was just short of what I needed (which caused a crash) hence
increase to 96kB (should probably be smarter about that).
commit 74aa739138d981311ce13076388382b5e89c6562
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:53:29 2012 +0100
gallivm: simplify aos float tex wrap repeat nearest
just handle pot and npot the same. The previous pot handling
ended up with exactly the same instructions plus 2 more (leave it
in the soa path though since it is probably still cheaper there).
While here also fix a issue which would cause a crash after an assert.
commit 0e1e755645e9e49cfaa2025191e3245ccd723564
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:29:24 2012 +0100
gallivm: (trivial) skip floor rounding in ifloor when not signed
This was only done for the non-sse41 case before, but even with
sse41 this is obviously unnecessary (some callers already call
itrunc in this case anyway but some might not).
commit 7f01a62f27dcb1d52597b24825931e88bae76f33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 25 11:23:12 2012 +0100
gallivm: (trivial) fix bogus comments
commit 5c85be25fd82e28490274c468ce7f3e6e8c1d416
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Jun 20 11:51:57 2012 +0100
translate: Free elt8_func/elt16_func too.
These were leaking.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 0ad498f36fb6f7458c7cffa73b6598adceee0a6c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 15:55:34 2012 +0200
gallivm: fix bug for tex wrap repeat with linear sampling in aos float path
The comparison needs to be against length not length_minus_one, otherwise
the max texel is never chosen (for the second coordinate).
Fixes piglit texwrap-1D-npot-proj (and 2D/3D versions).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit d1ad65937c5b76407dc2499b7b774ab59341209e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Jun 19 16:13:43 2012 +0200
gallivm: simplify soa tex wrap repeat with npot textures and no mip filtering
Similar to what is already done in aos sampling for the float path (but not
the int path since we don't get normalized float coordinates there).
URem is expensive and the calculation is done trivially with
normalized floats instead (at least with sse41-capable cpus).
(Some day should probably do the same for the mip filter path but it's much
more complicated there hence the gain is smaller.)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit e1e23f57ba9b910295c306d148f15643acc3fc83
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:38:56 2012 +0200
llvmpipe: (trivial) remove duplicated function declaration
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 07ca57eb09e04c48a157733255427ef5de620861
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 20:37:34 2012 +0200
llvmpipe: destroy setup variants on context destruction
lp_delete_setup_variants() used to be called in garbage collection,
but this no longer exists hence the setup shaders never got freed.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit ed0003c633859a45f9963a479f4c15ae0ef1dca3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 18 16:25:29 2012 +0100
gallivm: handle different ilod parts for multiple quad sampling
This fixes filtering when the integer part of the lod is not the same
for all quads. I'm not fully convinced of that solution yet as it just
splits the vector if the levels to be sampled from are different.
But otherwise we'd need to do things like some minify steps, and getting
mip level base address separately anyway hence it wouldn't really look
like much of a win (and making the code even more complex).
This should now give identical results to single quad sampling.
commit 8580ac4cfc43a64df55e84ac71ce1a774d33c0d2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:14:47 2012 +0200
gallivm: de-duplicate sample code common to soa and aos sampling
There doesn't seem to be any reason why this code dealing with cube face
selection, lod and mip level calculation is separate in aos and
soa sampling, and I am sick of having it to change in both places.
commit fb541e5f957408ce305b272100196f1e12e5b1e8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Jun 14 18:15:41 2012 +0200
gallivm: do mip filtering with per quad lod_fpart
This gives better results for mip filtering, though the generated code might
not be optimal. For now it also creates some artifacts if the lod_ipart isn't
the same for all quads, since instead of using the same mip weight for all
quads as previously (which just caused non-smooth gradients) this now will
use the right weights but with the wrong mip level in this case (can easily
be seen with things like texfilt, mipmap_tunnel).
v2: use logic helper suggested by José, and fix issue with negative lod_fpart
values
commit f1cc84eef7d826a20fab6cd8ccef9a275ff78967
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Jun 13 18:35:25 2012 +0200
gallivm: (trivial) fix bogus assert in lp_build_unpack_broadcast_aos_scalars
commit 7c17dbae8ae290df9ce0f50781a09e8ed640c044
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:11:14 2012 +0100
util: Reimplement half <-> float conversions.
Removed u_half.py used to generate the table for previous method.
Previous implementation of float to half conversion was faulty for
denormalised and NaNs and would require extra logic to fix,
thus making the speedup of using tables irrelevant.
commit 7762f59274070e1dd4b546f5cb431c2eb71ae5c3
Author: James Benton <jbenton@vmware.com>
Date: Tue Jun 12 12:12:16 2012 +0100
tests: Updated tests to properly handle NaN for half floats.
commit fa94c135aea5911fd93d5dfb6e6f157fb40dce5e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:33:10 2012 +0200
gallivm: do mip level calculations per quad
This is the final piece which shouldn't change the rendering output yet.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 23cbeaddfe03c09ca18c45d28955515317ffcf4c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Jun 9 00:54:21 2012 +0200
gallivm: do per-quad cube face selection
Doesn't quite fix the piglit cubemap test (not sure why actually)
but doing per-quad face selection is doing the right thing and
definitely an improvement.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit abfb372b3702ac97ac8b5aa80ad1b94a2cc39d33
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Jun 11 18:22:59 2012 +0200
gallivm: do all lod calculations per quad
Still no functional change but lod is now converted to scalar after
lod calculations.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 519368632747ae03feb5bca9c655eccbc5b751b4
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:46:10 2012 +0100
gallivm: Added support for half-float to float conversion in lp_build_conv.
Updated various utility functions to support this change.
commit 135b4d683a4c95f7577ba27b9bffa4a6fbd2c2e7
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 16:02:46 2012 +0100
gallivm: Added function for half-float to float conversion.
Updated lp_build_format_aos_array to support half-float source.
commit 37d648827406a20c5007abeb177698723ed86673
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:55:18 2012 +0100
util: Updated u_format_tests to rigidly test half-float boundary values.
commit 2ad18165d96e578aa9046df7c93cb1c3284d8c6b
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:54:16 2012 +0100
llvmpipe: Updated lp_test_format to properly handle Inf/NaN results.
commit 78740acf25aeba8a7d146493dd5c966e22c27b73
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 14:53:30 2012 +0100
util: Added functions for checking NaN / Inf for double and half-floats.
commit 35e9f640ae01241f9e0d67fe893bbbf564c05809
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:05:13 2012 +0200
gallivm: Fix calculating rho for 3d textures for the single-quad case
Discovered by accident, this looks like a very old typo bug.
commit fc1220c636326536fd0541913154e62afa7cd1d8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 24 21:04:59 2012 +0200
gallivm: do calcs per-quad in lp_build_rho
Still convert to scalar at the end of the function.
commit 50a887ffc550bf310a6988fa2cea5c24d38c1a41
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 21 23:21:50 2012 +0200
gallivm: (trivial) return scalar in lp_build_extract_range for length 1 vectors
Our type system on top of llvm's one doesn't generally support vectors of
length 1, instead using scalars. So we should return a scalar from this
function instead of having to bitcast the vector with length 1 later elsewhere.
commit 80c71c621f9391f0f9230460198d861643324876
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 17:49:15 2012 +0100
draw: Fixed bad merge error
commit c47401cfad0c9167de20ff560654f533579f452c
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:29:30 2012 +0100
draw: Updated store_clip to store whole vectors instead of individual elements.
commit 2d9c1ad74b0b0b41861fffcecde39f09cc27f1cf
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:28:32 2012 +0100
gallivm: Added lp_build_fetch_rgba_aos_array.
A version of lp_build_fetch_rgba_aos which is targeted at simple array formats.
Reads the whole vector from memory in one, instead of reading each element
individually.
Tested with mesa tests and demos.
commit ff7805dc2b6ef6d8b11ec4e54aab1633aef29ac8
Author: James Benton <jbenton@vmware.com>
Date: Tue May 22 15:27:40 2012 +0100
gallivm: Added lp_build_pad_vector.
This function pads a vector with undef to a desired length.
commit 701f50acef24a2791dabf4730e5b5687d6eb875d
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 17:27:19 2012 +0100
util: Added util_format_is_array.
This function checks whether a format description is in a simple array format.
commit 5e0a7fa543dcd009de26f34a7926674190fa6246
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:13:47 2012 +0100
draw: Removed draw_llvm_translate_from and draw/draw_llvm_translate.c.
This is "replaced" by adding an optimised path in lp_build_fetch_rgba_aos
in an upcoming patch.
commit 8c886d6a7dd3fb464ecf031de6f747cb33e5361d
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 15:02:31 2012 +0100
draw: Modified store_aos to write the vector as one, not individual elements.
commit 37337f3d657e21dfd662c7b26d61cb0f8cfa6f17
Author: James Benton <jbenton@vmware.com>
Date: Wed May 16 14:16:23 2012 +0100
draw: Changed aos_to_soa to use lp_build_transpose_aos.
commit bd2b69ce5d5c94b067944d1dcd5df9f8e84548f1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 19:14:27 2012 +0100
draw: Changed soa_to_aos to use lp_build_transpose_aos.
commit 0b98a950d29a116e82ce31dfe7b82cdadb632f2b
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:45 2012 +0100
gallivm: Added lp_build_transpose_aos which converts between aos and soa.
commit 69ea84531ad46fd145eb619ed1cedbe97dde7cb5
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 18:57:01 2012 +0100
gallivm: Added lp_build_interleave2_half aimed at AVX unpack instructions.
commit 7a4cb1349dd35c18144ad5934525cfb9436792f9
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 22 11:54:14 2012 +0100
gallivm: Fix build on Windows.
MC-JIT not yet supported there.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit afd105fc16bb75d874e418046b80d9cc578818a1
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:17:26 2012 +0100
llvmpipe: Added a error counter to lp_test_conv.
Useful for keeping track of progress when fixing errors!
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit b644907d08c10a805657841330fc23db3963d59c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:16:46 2012 +0100
llvmpipe: Changed known failures in lp_test_conv.
To comply with the recent fixes to lp_bld_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit d7061507bd94f6468581e218e61261b79c760d4f
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:14:38 2012 +0100
llvmpipe: Added fixed point types tests to lp_test_conv.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 146b3ea39b4726dbe125ac666bd8902ea3d6ca8c
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:26:35 2012 +0100
llvmpipe: Changed lp_test_conv src/dst alignment to be correct.
Now based on the define rather than a fixed number.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit f3b57441f834833a4b142a951eb98df0aa874536
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:06:44 2012 +0100
gallivm: Fixed erroneous optimisation in lp_build_min/max.
Previously assumed normalised was 0 to 1, but it can be -1 to 1
if type is signed.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a0613382e5a215cd146bb277646a6b394d376ae4
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:04:49 2012 +0100
gallivm: Compensate for lp_const_offset in lp_build_conv.
Fixing a /*FIXME*/ to remove errors in integer conversion in lp_build_conv.
Tested using lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit a3d2bf15ea345bc8a0664f8f441276fd566566f3
Author: James Benton <jbenton@vmware.com>
Date: Fri May 18 16:01:25 2012 +0100
gallivm: Fixed overflow in lp_build_clamped_float_to_unsigned_norm.
Tested with lp_test_conv and lp_test_format, reduced errors.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit e7b1e76fe237613731fa6003b5e1601a2e506207
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 21 20:07:51 2012 +0100
gallivm: Fix build with LLVM 2.6
Trivial, and useful.
commit d3c6bbe5c7f5ba1976710831281ab1b6a631082d
Author: José Fonseca <jfonseca@vmware.com>
Date: Tue May 15 17:15:59 2012 +0100
gallivm: Enable MCJIT/AVX with vanilla LLVM 3.1.
Add the necessary C++ glue, so that we don't need any modifications
to the soon to be released LLVM 3.1.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
commit 724a019a14d40fdbed21759a204a2bec8a315636
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 22:04:06 2012 +0100
gallivm: Use HAVE_LLVM 0x0301 consistently.
commit af6991e2a3868e40ad599b46278551b794839748
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon May 14 21:49:06 2012 +0100
gallivm: Add MCRegisterInfo.h to silence benign warnings about missing implementation.
Trivial.
commit 6f8a1d75458daae2503a86c6b030ecc4bb494e23
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Apr 2 22:14:15 2012 -0700
gallivm: Pass in a MCInstrInfo to createMCInstPrinter on llvm-3.1.
llvm-3.1svn r153860 makes MCInstrInfo available to the MCInstPrinter.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 62555b6ed8760545794f83064e27cddcb3ce5284
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 27 21:51:17 2012 -0700
gallivm: Fix method overriding in raw_debug_ostream.
Use matching type qualifers to avoid method hiding.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit 6a9bd784f4ac68ad0a731dcd39e5a3c39989f2be
Author: Vinson Lee <vlee@freedesktop.org>
Date: Tue Mar 13 22:40:52 2012 -0700
gallivm: Fix createOProfileJITEventListener namespace with llvm-3.1.
llvm-3.1svn r152620 refactored the OProfile profiling code.
createOProfileJITEventListener was moved from the llvm namespace to the
llvm::JITEventListener namespace.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit b674955d39adae272a779be85aa1bd665de24e3e
Author: Vinson Lee <vlee@freedesktop.org>
Date: Mon Mar 5 22:00:40 2012 -0800
gallivm: Pass in a MCRegisterInfo to MCInstPrinter on llvm-3.1.
llvm-3.1svn r152043 changes createMCInstPrinter to take an additional
MCRegisterInfo argument.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 11ab69971a8a31c62f6de74905dbf8c02884599f
Author: Vinson Lee <vlee@freedesktop.org>
Date: Wed Feb 29 21:20:53 2012 -0800
Revert "gallivm: Change getExtent and readByte to non-const with llvm-3.1."
This reverts commit d5a6c172547d8964f4d4bb79637651decaf9deee.
llvm-3.1svn r151687 makes MemoryObject accessor members const again.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 339960c82d2a9f5c928ee9035ed31dadb7f45537
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon May 14 16:19:56 2012 +0200
gallivm: (trivial) fix assertion failure for mipmapped 1d textures
In lp_build_rho, we may end up with a 1-element vector (for mipmapped 1d
textures), but in this case we require the type to be a non-vector type,
so need a cast.
commit 9d73edb727bd6d196030dc3026b7bf0c574b3e19
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:12:07 2012 +0200
gallivm: prepare for per-quad lod calculations for large vectors
to be able to handle multiple quads at once in texture sampling and still
do lod calculations per quad, it is necessary to get the per-quad derivatives
into the lp_build_rho function.
Until now these derivative values were just scalars, which isn't going to work.
So we now use vectors, and since the interface needs to change we also do some
different (slightly more efficient) packing of the values.
For 8-wide vectors the packed derivative values for 3 coords would look like
this, this scales to a arbitrary (multiple of 4) vector size:
ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
dr1dx dr1dy _____ _____ dr2dx dr2dy _____ _____
The second vector will be unused for 1d and 2d textures.
To facilitate future changes the derivative values are put into a struct, since
quite some functions just pass these values through.
The generated code seems to be very slightly better for 2d textures (with
4-wide vectors) than before with sse2 (if you have a cpu with physical 128bit
simd units - otherwise it's probably not a win).
v2: suggestions from José, rename variables, add comments, use swizzle helper
commit 0aa21de0d31466dac77b05c97005722e902517b8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu May 10 18:10:31 2012 +0200
gallivm: add undefined swizzle handling to lp_build_swizzle_aos
This is useful for vectors with "holes", it lets llvm choose the most
efficient shuffle instructions if some elements aren't needed without having to
worry what elements to manually pick otherwise.
commit 00faf3f370e7ce92f5ef51002b0ea42ef856e181
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri May 4 17:25:16 2012 +0100
gallivm: Get the LLVM IR optimization passes before JIT compilation.
MC-JIT engine compiles the module immediately on creation, so the optimization
passes were being run too late.
So now we create a target data layout from a string, that matches the
ABI parameters reported by the compiler.
The backend optimization passes were always been run, so the performance
improvement is modest (3% on multiarb mesa demo).
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
commit 40a43f4e2ce3074b5ce9027179d657ebba68800a
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed May 2 16:03:54 2012 +0200
gallivm: (trivial) fix wrong define used in lp_build_pack2
should fix stack-smashing crashes.
commit e6371d0f4dffad4eb3b7a9d906c23f1c88a2ab9e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 30 21:25:29 2012 +0200
gallivm: add perf warnings when not using intrinsics with 256bit vectors
Helper functions using integer sse2 intrinsics could split the vectors with AVX
instead of using generic fallback (which should be faster).
We don't actually expect to hit these paths (hence don't fix them up to actually
do the vector splitting) so just emit warnings (for those functions where it's
obvious doing split/intrinsic is faster than using generic path).
Only emit warnings for 256bit vectors since we _really_ don't expect to hit
arbitrary large vectors which would affect a lot more functions.
The warnings do not actually depend on avx since the same logic applies to
plain sse2 too (but of course again there's _really_ no reason we should hit
these functions with 256bit vectors without avx).
commit 8a9ea701ea7295181e846c6383bf66a5f5e47637
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:37:07 2012 +0200
gallivm: split vectors manually for avx in lp_build_pack2 (v2)
There's 2 reasons for this:
First, there's a llvm bug (fixed in 3.1) which generates tons of byte
inserts/extracts otherwise, and second, more importantly, we want to use
pack intrinsics instead of shuffles.
We do this in lp_build_pack2 and not the calling code (aos sample path)
because potentially other callers might find that useful too, even if
for larger sequences of code using non-native vector sizes it might be
better to manually split vectors.
This should boost texture performance in the aos path considerably.
v2: fix issues with intrinsics types with old llvm
commit 27ac5b48fa1f2ea3efeb5248e2ce32264aba466e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:26:22 2012 +0200
llvmpipe: refactor lp_build_pack2 (v2)
prettify, and it's unnecessary to assert when there's no intrinsic due to
unsupported bit width - the shuffle path will work regardless.
In contrast lp_build_packs2, should only rely on lp_build_pack2 doing the
clamping for element sizes for which there is a sse2 intrinsic.
v2: fix bug spotted by Jose regarding the intrinsic type for packusdw
on old llvm versions.
commit ddf279031f0111de4b18eaf783bdc0a1e47813c8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue May 1 20:13:59 2012 +0200
gallivm: add src width check in lp_build_packs2()
not doing so would skip clamping even if no sse2 pack instruction is
available, which is incorrect (in theory only, such widths would also always
hit a (unnecessary) assertion in lp_build_pack2().
commit e7f0ad7fe079975eae7712a6e0c54be4fae0114b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Apr 27 15:57:00 2012 +0200
gallivm: (trivial) fix crash-causing typo for npot textures with avx
commit 28a9d7f6f655b6ec508c8a3aa6ffefc1e79793a0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Apr 25 19:38:45 2012 +0200
gallivm: (trivial) remove code mistakenly added twice.
commit d5926537316f8ff67ad0a52e7242f7c5478d919b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Apr 24 21:16:15 2012 +0200
gallivm: add a new avx aos sample path (v2)
Try to avoid mixing float and int address calculations. This does texture wrap
modes with floats, and then the offset calculations still with ints (because
of lack of precision with floats, though we could do some effort to make it work
with not too large (16MB) textures).
This also handles wrap repeat mode with npot-sized textures differently than
either the old soa or aos int path (likely way faster but untested).
Otherwise the actual address wrap code is largely similar to the soa path (not
quite the same as this one also has some int code), it should get used by avx
soa sampling later as well but doesn't handle more complex address modes yet
(this will also have the benefit that we can use aos sampling path for all
texture address modes).
Generated code for that looks reasonable, but still does not split vectors
explicitly for fetch/filter which means still get hit by llvm (fixed upstream)
which generates hundreds of pinsrb/pextrb instead of two shuffles.
It is not obvious though if it's much of a win over just doing address calcs
4-wide but with ints, even if it is definitely much less instructions on avx.
piglit's texwrap seems to look exactly the same but doesn't test
neither the non-normalized nor the npot cases.
v2: fix comments, prettify based on Brian's and Jose's feedback.
commit bffecd22dea66fb416ecff8cffd10dd4bdb73fce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Apr 19 01:58:29 2012 +0200
gallivm: refactor aos lp_build_sample_image_nearest/linear
split them up to separate address calculations and fetching/filtering.
Need this for being able to do 8-wide float address calcs and 4-wide
fetch/filter later (for avx). Plus the functions were very big scary monsters
anyway (in particular lp_build_sample_image_linear).
commit a80b325c57529adddcfa367f96f03557725c4773
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Apr 16 17:17:18 2012 +0200
gallivm: fix lp_build_resize when truncating width but expanding vector size
Missed this case which I thought was impossible - the assertion for it was
right after the division by zero...
(AoS) texture sampling may ask us to do this, for things like 8 4x32int
vectors to 1 32x8int vector conversion (eventually, we probably don't want
this to happen).
commit f9c8337caa3eb185830d18bce8b95676a065b1d7
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sat Apr 14 18:00:59 2012 +0200
gallivm: fix cube maps with larger vectors
This makes the branchless cube face selection code work with larger vectors.
Because the complexity is quite high (cannot really be improved it seems,
per-face selection would reduce complexity a lot but this leads to errors
unless the derivatives are calculated all from the same face which almost
doubles the work to be done) it is still slower than the branching version,
hence only enable this with large vectors.
It doesn't actually do per-quad face selection yet (only makes sense with
matching lod selection, in fact it will select the same face for all pixels
based on the average of the first four pixels for now) but only different
shuffles are required to make it work (the branching version actually should
work with larger vectors too now thanks to the improved horizontal add but of
course it cannot be extended to really select the face per-quad unless doing
branching per quad).
commit 7780c58869fc9a00af4f23209902db7e058e8a66
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 21:11:12 2012 +0100
llvmpipe: (trivial) fix compiler warning
and also clarify comment regarding availability of popcnt instruction.
commit a266dccf477df6d29a611154e988e8895892277e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 30 14:21:07 2012 +0100
gallivm: remove unneeded members in lp_build_sample_context
Minor cleanup, the texture width, height, depth aren't accessed in their
scalar form anywhere. Makes it more obvious those values should probably be
fetched already vectorized (but this requires more invasive changes)...
commit b678c57fb474e14f05e25658c829fc04d2792fff
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 29 15:53:55 2012 +0100
gallivm: add a helper for concatenating vectors
Similar to the extract_range helper intended to get around slow code generated
by llvm for 128bit insertelements.
Concatenating two 128bit vectors this way will result in a single vinsertf128
operation rather than two 64bit stores plus one 128bit load, though it might be
mildly useful for other purposes as well.
commit 415ff228bcd0cf5e44a4c15350a661f0f5520029
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 19:41:15 2012 +0100
gallivm: add a custom 2x8f->1x16ub avx conversion path
Similar to the existing 4x4f->1x16ub sse2 path, shaves off a couple
instructions (min/max mostly) because it relies on pack intrinsics clamping.
commit 78c08fc89f8fbcc6dba09779981b1e873e2a0299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 18:44:07 2012 +0100
gallivm: add avx arithmetic intrinsics
Add all avx intrinsics for arithmetic functions (with the exception
of the horizontal add function which needs another look).
Seems to pass basic tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
commit a586caa2800aa5ce54c173f7c0d4fc48153dbc4e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 15:31:35 2012 +0100
gallivm: add avx logic intrinsics
Add the blend intrinsics for 8-wide float and 4-wide double vectors.
Since we lack 256bit int instructions these are used for int vectors as well,
though obviously not for byte or word element values.
The comparison intrinsics aren't extended for avx since these are only used
for pre-2.7 llvm versions.
commit 70275e4c13c89315fc2560a4c488c0e6935d5caf
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 28 00:40:53 2012 +0100
gallivm: new helper function for extract shuffles.
Based on José's idea as we can need that in a couple places.
Note that such shuffles should not be used lightly, since data layout
of <4 x i8> is different to <16 x i8> for instance, hence might cause
data rearrangement.
commit 4d586dbae1b0c55915dda1759d2faea631c0a1c2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 18:27:25 2012 +0100
gallivm: (trivial) don't overallocate shuffle variable
using wrong define meant huge array...
commit 06b0ec1f6d665d98c135f9573ddf4ba04b2121ad
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 27 17:54:20 2012 +0100
gallivm: don't do per-element extract/insert for vector element resize
Instead of doing per-element extract/insert if the src vectors
and dst vector differ in total size (which generates atrocious code)
first change the src vectors size by using shuffles to destination
vector size.
We can still do better than that on AVX for packing to color buffer
(by exploiting pack intrinsics characteristics hence eleminating the
need for some clamps) but this already generates much better code.
v2: incorporate feedback from José, Keith and use shuffle instead of
bitcasts/extracts. Due to llvm deficiencies the latter cause all data
to get moved to GPRs and back in pieces (even though the data in the
regs actually stays the same...).
commit c9970d70e05f95d3f52fe7d2cd794176a52693aa
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 19:33:19 2012 +0000
gallivm: fix bug in simple position interpolation
Accidental use of position attribute instead of just pixel coordinates.
Caused failures in piglit glsl-fs-ceil and glsl-fs-floor.
commit d0b6fcdb008d04d7f73d3d725615321544da5a7e
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 23 15:31:14 2012 +0000
gallivm: fix emission of ceil opcode
lp_build_ceil seems more appropriate than lp_build_trunc.
This seems to be never hit though someone performs some ceil
to floor magic.
commit d97fafed7e62ffa6bf76560a92ea246a1a26d256
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 22 11:46:52 2012 +0000
gallivm: new vectorized path for cubemap calculations
should be faster when adapted to multiple quads as only selection masks need to be different.
The code is more or less a per-pixel version adapted to only do it per quad.
A per pixel version would be much simpler (could drop 2 selects, 6 broadcasts and the messy
horizontal add of 3 vectors at the expense of only 2 more absolute value instructions -
would also just work for arbitary large vectors).
This version doesn't yet work with larger vectors because the horizontal add isn't adjusted
to be able to work with 2x4 vectors (and also because face selection wouldn't be done per
quad just per block though that would be only a correctness issue just as with lod selection).
The downside is this code is quite a bit slower. On a Core2 it can be sped up by disabling the
hw blend instructions for selection and using logicop fallbacks instead, but it is still slower
than the old code, hence leave that in for now. Probably will chose one or the other version
based on vector length in the end.
commit b375fbb18a3fd46859b7fdd42f3e9908ea4ff9a3
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 21 14:42:29 2012 +0000
gallivm: fix optimized occlusion query intrinsic name
commit a9ba0a3b611e48efbb0e79eb09caa85033dbe9a2
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Mar 21 16:19:43 2012 +0000
draw,gallivm,llvmpipe: Call gallivm_verify_function everywhere.
commit f94c2238d2bc7383e088b8845b7410439a602071
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 18:54:10 2012 +0000
gallivm: optimize calculations for cube maps a bit
this does some more vectorized calculations and uses horizontal adds if possible.
A definite win with sse3 otherwise it doesn't seem to make much of a difference.
In any case this is arithmetically identical, cannot handle larger vectors.
Should be useful as a reference point against larger vector version later...
commit 21a2c1cf3c8e1ac648ff49e59fdc0e3be77e2ebb
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 15:16:27 2012 +0000
llvmpipe: slight optimization of occlusion queries
using movmskps when available.
While this is slightly better for cpus without popcnt we should
really sum the vectors ourselves (it is also possible to cast to i4 before
doing the popcnt but that doesn't help that much neither since llvm
is using some optimized popcnt version for i32)
commit 5ab5a35f216619bcdf55eed52b0db275c4a06c1b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 20 13:32:11 2012 +0000
llvmpipe: fix occlusion queries with larger vectors
need to adjust casts etc.
commit ff95e6fdf5f16d4ef999ffcf05ea6e8c7160b0d5
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Mar 19 20:15:25 2012 +0000
gallivm: Restore optimization passes.
commit 57b05b4b36451e351659e98946dae27be0959832
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:34:22 2012 +0000
llvmpipe: use existing min2 macro
commit bc9a20e19b4f600a439f45679451f2e87cd4b299
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 19:07:27 2012 +0000
llvmpipe: add some safeguards against really large vectors
As per José's suggestion, prevent things from blowing up if some cpu
would have 1024bit or larger vectors.
commit 0e2b525e5ca1c5bbaa63158bde52ad1c1564a3a9
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:31:08 2012 +0000
llvmpipe: fix mask generation for uberwide vectors
this was the only piece preventing 16-wide vectors from working
(apart from the LP_MAX_VECTOR_WIDTH define that is), which is the maximum
as we don't get more pixels in the fragment shader at once.
Hence adjust that so things could be tested properly with that size
even though there seems to be no practical value.
commit 3c8334162211c97f3a11c7f64e9e5a2a91ad9656
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 18:19:41 2012 +0000
llvmpipe: fix the simple interpolation method with larger vectors
so both methods actually _really_ work now. Makes textures look
nice with larger vectors...
commit 1cb0464ef8871be1778d43b0c56adf9c06843e2d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 17:26:35 2012 +0000
llvmpipe: fix mask generation and position interpolation with 8-wide vectors
trivial bugs, with these things start to look somewhat reasonable.
Textures though have some swizzling issues it seems.
commit 168277a63ef5b72542cf063c337f2d701053ff4b
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 16:04:03 2012 +0000
llvmpipe: don't overallocate variables
we never have more than 16 (stamp size) / 4 (minimum possible vector size).
(With larger vectors those variables are still overallocated a bit.)
commit 409b54b30f81ed0aa9ed0b01affe15c72de9abd2
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:56:48 2012 +0000
llvmpipe: add some 32f8 formats to lp_test_conv
Also add the ability to handle different sized vectors.
commit 55dcd3af8366ebdac0af3cdb22c2588f24aa18ce
Author: Roland Scheidegger <sroland@vmware.com>
Date: Mon Mar 19 15:47:27 2012 +0000
gallivm: handle different sized vectors in conversion / pack
only fully generic path for now (extract/insert per element).
commit 9c040f78c54575fcd94a8808216cf415fe8868f6
Author: Roland Scheidegger <sroland@vmware.com>
Date: Sun Mar 18 00:58:28 2012 +0100
llvmpipe: fix harmless use of unitialized values
commit 551e9d5468b92fc7d5aa2265db9a52bb1e368a36
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:31:21 2012 +0100
gallivm: drop special path in extract_broadcast with different sized vectors
Not needed, llvm can handle shuffles with different sized result vector just
fine. Should hopefully generate the same code in the end, but simpler IR.
commit 44da531119ffa07a421eaa041f63607cec88f6f8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 16 23:28:49 2012 +0100
llvmpipe: adapt interpolation for handling multiple quads at once
this is still WIP there are actually two methods possible not quite
sure what makes the most sense, so there's code for both for now:
1) the iterative method as used before (compute attrib values at upper left
corner of stamp and upper left corner of each quad initially).
It is improved to handle more than one quad at once, and also do some more vectorized
calculations initially for slightly better code - newer cpus have full throughput with
4 wide float vectors, hence don't try to code up a path which might be faster if there's
just one channel active per attribute.
2) just do straight interpolation for each pixel.
Method 2) is more work per quad, but less initially - if all quads are executed
significantly more overall though. But this might change with larger vector lengths.
This method would also be needed if we'd do some kind of active quad merging when
operating on multiple quads at once.
This path contains some hack to force llvm to generate better code, it is still far
from ideal though, still generates far too many unnecessary register spills/reloads.
Both methods should work with different sized vectors.
Not very well tested yet, still seems to work with four-wide vectors, need changes
elsewhere to be able to test with wider vectors.
commit be5d3e82e2fe14ad0a46529ab79f65bf2276cd28
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:59:37 2012 +0000
draw: Cleanup.
commit f85bc12c7fbacb3de2a94e88c6cd2d5ee0ec0e8d
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 16 20:43:30 2012 +0000
gallivm: More module compilation refactoring.
commit d76f093198f2a06a93b2204857e6fea5fd0b3ece
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Mar 15 21:29:11 2012 +0000
llvmpipe: Use gallivm_compile/free_function() in linear code.
Should had been done before.
commit 122e1adb613ce083ad739b153ced1cde61dfc8c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 13 14:47:10 2012 +0100
llvmpipe: generate partial pixel mask for multiple quads
still works with one quad, cannot be tested yet with more
At least for now always fixed order with multiple quads.
commit 4c4f15081d75ed585a01392cd2dcce0ad10e0ea8
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 22:09:24 2012 +0100
llvmpipe: refactor state setup a bit
Refactor to make it easier to emit (and potentially later fetch in fs)
coefficients for multiple attributes at once.
Need to think more about how to make this actually happen however, the
problem is different attributes can have different interpolation modes,
requiring different handling in both setup and fs (though linear and
perspective handling is close).
commit 9363e49722ff47094d688a4be6f015a03fba9c79
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 19:23:23 2012 +0100
llvmpipe: vectorize tri offset calc
cuts number of instructions in quad-offset-factor from 107 to 75.
This code actually duplicated the (scalar) code calculating the determinant
except it used different vertex order (leading to different sign but it doesn't
matter) hence llvm could not have figured out it's the same (of course with
determinant vectorized in the other place that wouldn't have worked any longer
neither).
Note this particular piece doesn't actually vectorize well, not many arithmetic
instructions left but tons of shuffle instructions...
Probably would need to work on n tris at a time for better vectorization.
commit 63169dcb9dd445c94605625bf86d85306e2b4297
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Mar 8 03:11:37 2012 +0100
llvmpipe: vectorize some scalar code in setup
reduces number of arithmetic instructions, and avoids loading
vector x,y values twice (once as scalars once as vectors).
Results in a reduction of instructions from 76 to 64 in fs setup for glxgears
(16%) on a cpu with sse41.
Since this code uses vec2 disguised as vec4, on old cpus which had physical
64bit sse units (pre-Core2) it probably is less of a win in practice (and if
you have no vectors you can only hope llvm eliminates the arithmetic for
unneeded elements).
commit 732ecb877f951ab89bf503ac5e35ab8d838b58a1
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Mar 7 00:32:24 2012 +0100
draw: fix clipping
bug introduced by 4822fea3f0440b5205e957cd303838c3b128419c broke
clipping pretty badly (verified with lineclip test)
commit ef5d90b86d624c152d200c7c4056f47c3c6d2688
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 23:38:59 2012 +0100
draw: don't store vertex header per attribute
storing the vertex header once per attribute is totally unnecessary.
Some quick look at the generated assembly says llvm in fact cannot optimize
away the additional stores (maybe due to potentially aliasing pointers
somewhere).
Plus, this makes the code cleaner and also allows using a vector "or"
instead of scalar ones.
commit 6b3a5a57b0b9850854cfbd7b586e4e50102dda71
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Mar 6 19:11:01 2012 +0100
draw: do the per-vertex "boolean" clipmask "or" with vectors
no point extracting the values and doing it per component.
Doesn't help that much since we still extract the values elsewhere anyway.
commit 36519caf1af40e4480251cc79a2d527350b7c61f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Fri Mar 2 22:27:01 2012 +0100
gallivm: fix lp_build_extract_broadcast with different sized vectors
Fix the obviously wrong argument, so it doesn't blow up.
commit 76d0ac3ad85066d6058486638013afd02b069c58
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Mar 2 12:16:23 2012 +0000
draw: Compile per module and not per function (WIP).
Enough to get gears w/ LLVM draw + softpipe to work on AVX doing:
GALLIUM_DRIVER=softpipe SOFTPIPE_USE_LLVM=yes glxgears
But still hackish -- will need to rethink and refactor this.
commit 78e32b247d2a7a771be9a1a07eb000d1e54ea8bd
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 12:01:05 2012 +0000
llvmpipe: Remove lp_state_setup_fallback.
Never used.
commit 6895d5e40d19b4972c361e8b83fdb7eecda3c225
Author: José Fonseca <jfonseca@vmware.com>
Date: Mon Feb 27 19:14:27 2012 +0000
llvmpipe: Don't emit EMMS on x86
We already take precautions to ensure that LLVM never emits MMX code.
commit 4822fea3f0440b5205e957cd303838c3b128419c
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 29 15:58:19 2012 +0100
draw: modifications for larger vector sizes
We want to be able to use larger vectors especially for running the vertex
shader. With this patch we build soa vectors which might have a different
length than 4.
Note that aos structures really remain the same, only when aos structures
are converted to soa potentially different sized vectors are used.
Samplers probably don't work yet, didn't look at them.
Testing done:
glxgears works with both 128bit and 256bit vectors.
commit f4950fc1ea784680ab767d3dd0dce589f4e70603
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:57 2012 +0100
gallivm: override native vector width with LP_NATIVE_VECTOR_WIDTH env var for debug
commit 6ad6dbf0c92f3bf68ae54e5f2aca035d19b76e53
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 29 15:51:24 2012 +0100
draw: allocate storage with alignment according to native vector width
commit 7bf0e3e7c9bd2469ae7279cabf4c5229ae9880c1
Author: José Fonseca <jfonseca@vmware.com>
Date: Fri Feb 24 19:06:08 2012 +0000
gallivm: Fix comment grammar.
Was missing several words. Spotted by Roland.
commit b20f1b28eb890b2fa2de44a0399b9b6a0d453c52
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 19:22:09 2012 +0000
gallivm: Use MC-JIT on LLVM 3.1 + (i.e, SVN)
MC-JIT
Note: MC-JIT is still WIP. For this to work correctly it requires
LLVM changes which are not yet upstream.
commit b1af4dfcadfc241fd4023f4c3f823a1286d452c0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 20:03:15 2012 +0100
llvmpipe: use new lp_type_width() helper in lp_test_blend
commit 04e0a37e888237d4db2298f31973af459ef9c95f
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 19:50:34 2012 +0100
llvmpipe: clean up lp_test_blend a little
Using variables just sized and aligned right makes it a bit more obvious
what's going on.
The test still only tests vector length 4.
For AoS anything else probably isn't going to work.
For SoA other lengths should work (at least with floats).
commit e61c393d3ec392ddee0a3da170e985fda885a823
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:48:30 2012 +0000
gallivm: Ensure vector width consistency.
Instead of assuming that everything is the max native size.
commit 330081ac7bc41c5754a92825e51456d231bf84dd
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:44:14 2012 +0000
draw: More simd vector width consistency fixes.
commit d90ca002753596269e37297e2e6c139b19f29f03
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 17:43:00 2012 +0000
gallivm: Remove unused lp_build_int32_vec4_type() helper.
commit cae23417824d75869c202aaf897808d73a2c1db0
Author: Roland Scheidegger <sroland@vmware.com>
Date: Thu Feb 23 17:32:16 2012 +0100
gallivm: use global variable for native vector width instead of define
We do not know the simd extensions (and hence the simd width we should use)
available at compile time.
At least for now keep a define for maximum vector width, since a global
variable obviously can't be used to adjust alignment of automatic stack
variables.
Leave the runtime-determined value at 128 for now in all cases.
commit 51270ace6349acc2c294fc6f34c025c707be538a
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 15:41:02 2012 +0000
gallivm: Add a hunk inadvertedly lost when rebasing.
commit bf256df9cfdd0236637a455cbaece949b1253e98
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:24:23 2012 +0000
llvmpipe: Use consistent vector width in depth/stencil test.
commit 5543b0901677146662c44be2cfba655fd55da94b
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 14:19:59 2012 +0000
draw: Use a consistent the vector register width.
Instead of 4x32 sometimes, LP_NATIVE_VECTOR_WIDTH other times.
commit eada8bbd22a3a61f549f32fe2a7e408222e5c824
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 12:08:04 2012 +0000
gallivm: Remove garbagge collection.
MC-JIT will require one compilation per module (as opposed to one
compilation per function), therefore no state will be shared,
eliminating the need to do garbagge collection.
commit 556697ea0ed72e0641851e4fbbbb862c470fd7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 10:33:41 2012 +0000
gallivm: Move all native target initialization to lp_set_target_options().
commit c518e8f3f2649d5dc265403511fab4bcbe2cc5c8
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:52:32 2012 +0000
llvmpipe: Create one gallivm instance for each test.
commit 90f10af8920ec6be6f2b1e7365cfc477a0cb111d
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:48:08 2012 +0000
gallivm: Avoid LLVMAddGlobalMapping() in lp_bld_assert().
Brittle, complex, and unecesary. Just use function pointer constant.
commit 98fde550b33401e3fe006af59db4db628bcbf476
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:21:26 2012 +0000
gallivm: Add a lp_build_const_func_pointer() helper.
To be reused in all places where we want to call C code.
commit 6cfedadb62c2ce5af8d75969bc95a607f3ece118
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 09:44:41 2012 +0000
gallivm: Cleanup/simplify lp_build_const_string_variable.
- Move to lp_bld_const where it belongs
- Rename to lp_build_const_string
- take the length from the argument (and don't count the zero terminator twice)
- bitcast the constant to generic i8 *
commit db1d4018c0f1fa682a9da93c032977659adfb68c
Author: José Fonseca <jfonseca@vmware.com>
Date: Thu Feb 23 11:52:17 2012 +0000
gallivm: Set NoFramePointerElimNonLeaf to true where supported.
commit 088614164aa915baaa5044fede728aa898483183
Author: Roland Scheidegger <sroland@vmware.com>
Date: Wed Feb 22 19:38:47 2012 +0100
llvmpipe: pass in/out pointers rather scalar floats in lp_bld_arit
we don't want llvm to potentially optimize away the vectors (though it doesn't
seem to currently), plus we want to be able to handle in/out vectors of arbitrary
length.
commit 3f5c4e04af8a7592fdffa54938a277c34ae76b51
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:22:55 2012 +0100
gallivm: fix lp_build_sqrt() for vector length 1
since we optimize away vectors with length 1 need to emit intrinsic
without vector type.
commit 79d94e5f93ed8ba6757b97e2026722ea31d32c06
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 17:00:46 2012 +0000
llvmpipe: Remove lp_test_round.
commit 81f41b5aeb3f4126e06453cfc78990086b85b78d
Author: Roland Scheidegger <sroland@vmware.com>
Date: Tue Feb 21 23:56:24 2012 +0100
llvmpipe: subsume lp_test_round into lp_test_arit
Much simpler, and since the arguments aren't passed as 128bit values can run
on any arch.
This also uses the float instead of the double versions of the c functions
(which probably was the intention anyway).
In contrast to lp_test_round the output is much less verbose however.
Tested vector width of 32 to 512 bits - all pass except 32 (length 1) which
crashes in lp_build_sqrt() due to wrong type.
Signed-off-by: José Fonseca <jfonseca@vmware.com>
commit 945b338b421defbd274481d8c4f7e0910fd0e7eb
Author: José Fonseca <jfonseca@vmware.com>
Date: Wed Feb 22 09:55:03 2012 +0000
gallivm: Centralize the function compilation logic.
This simplifies a lot of code.
Also doing this in a central place will make it easier to carry out the
changes necessary to use MC-JIT in the future.
gallivm: Fix typo in explicit derivative shuffle.
Trivial.
draw: make DEBUG_STORE work again
adapt to lp_build_printf() interface changes
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: get rid of vecnf_from_scalar()
just use lp_build_broadcast directly (cannot assign a name but don't really
need it, vecnf_from_scalar() was producing much uglier IR due to using
repeated insertelement instead of insertelement+shuffle).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
llvmpipe: fix typo in complex interpolation code
Fixes position interpolation when using complex mode
(piglit fp-fragment-position and similar)
Reviewed-by: José Fonseca <jfonseca@vmware.com>
draw: fix clipvertex/position storing again
This appears to be the result of a bad merge.
Fixes piglit tests relying on clipping, like a lot of the interpolation tests.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Fix explicit derivative manipulation.
Same counter variable was being used in two nested loops. Use more
meanigful variable names for the counter to fix and avoid this.
gallivm: Prevent buffer overflow in repeat wrap mode for NPOT.
Based on Roland's patch, discussion, and review .
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix dims for TGSI_TEXTURE_1D in emit_tex.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: Fix explicit volume texture derivatives.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
gallivm: fix 1d shadow texture sampling
Always r coordinate is used, hence need 3 coords not two
(the second one is unused).
Reviewed-by: José Fonseca <jfonseca@vmware.com>
gallivm: Enable AVX support without MCJIT, where available.
For now, this just enables AVX on Windows for testing. If the code is
stable then we might consider prefering the old JIT wherever possible.
No change elsewhere.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
2012-07-13 18:09:30 +01:00
|
|
|
|
2013-07-11 22:15:44 +01:00
|
|
|
/* rsqrt(1.0) != 1.0 here */
|
|
|
|
res = lp_build_fast_rsqrt(bld, a);
|
|
|
|
|
2012-09-21 16:03:48 +01:00
|
|
|
if (num_iterations) {
|
|
|
|
/*
|
|
|
|
* Newton-Raphson will result in NaN instead of infinity for zero,
|
|
|
|
* and NaN instead of zero for infinity.
|
|
|
|
* Also, need to ensure rsqrt(1.0) == 1.0.
|
|
|
|
* All numbers smaller than FLT_MIN will result in +infinity
|
|
|
|
* (rsqrtps treats all denormals as zero).
|
|
|
|
*/
|
|
|
|
LLVMValueRef cmp;
|
|
|
|
LLVMValueRef flt_min = lp_build_const_vec(bld->gallivm, type, FLT_MIN);
|
2015-03-18 14:19:10 +00:00
|
|
|
LLVMValueRef inf = lp_build_const_vec(bld->gallivm, type, INFINITY);
|
2010-08-14 18:02:47 +01:00
|
|
|
|
2012-09-21 16:03:48 +01:00
|
|
|
for (i = 0; i < num_iterations; ++i) {
|
|
|
|
res = lp_build_rsqrt_refine(bld, a, res);
|
|
|
|
}
|
|
|
|
cmp = lp_build_compare(bld->gallivm, type, PIPE_FUNC_LESS, a, flt_min);
|
|
|
|
res = lp_build_select(bld, cmp, inf, res);
|
|
|
|
cmp = lp_build_compare(bld->gallivm, type, PIPE_FUNC_EQUAL, a, inf);
|
|
|
|
res = lp_build_select(bld, cmp, bld->zero, res);
|
|
|
|
cmp = lp_build_compare(bld->gallivm, type, PIPE_FUNC_EQUAL, a, bld->one);
|
|
|
|
res = lp_build_select(bld, cmp, bld->one, res);
|
|
|
|
}
|
2010-08-14 18:02:47 +01:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
|
|
|
|
return lp_build_rcp(bld, lp_build_sqrt(bld, a));
|
|
|
|
}
|
|
|
|
|
2013-07-11 22:15:44 +01:00
|
|
|
/**
|
|
|
|
* If there's a fast (inaccurate) rsqrt instruction available
|
|
|
|
* (caller may want to avoid to call rsqrt_fast if it's not available,
|
|
|
|
* i.e. for calculating x^0.5 it may do rsqrt_fast(x) * x but if
|
|
|
|
* unavailable it would result in sqrt/div/mul so obviously
|
|
|
|
* much better to just call sqrt, skipping both div and mul).
|
|
|
|
*/
|
|
|
|
boolean
|
|
|
|
lp_build_fast_rsqrt_available(struct lp_type type)
|
|
|
|
{
|
|
|
|
assert(type.floating);
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if ((util_get_cpu_caps()->has_sse && type.width == 32 && type.length == 4) ||
|
|
|
|
(util_get_cpu_caps()->has_avx && type.width == 32 && type.length == 8)) {
|
2013-07-11 22:15:44 +01:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Generate 1/sqrt(a).
|
|
|
|
* Result is undefined for values < 0, infinity for +0.
|
|
|
|
* Precision is limited, only ~10 bits guaranteed
|
|
|
|
* (rsqrt 1.0 may not be 1.0, denorms may be flushed to 0).
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_fast_rsqrt(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
|
|
|
|
assert(lp_check_value(type, a));
|
|
|
|
|
|
|
|
if (lp_build_fast_rsqrt_available(type)) {
|
|
|
|
const char *intrinsic = NULL;
|
|
|
|
|
|
|
|
if (type.length == 4) {
|
|
|
|
intrinsic = "llvm.x86.sse.rsqrt.ps";
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
intrinsic = "llvm.x86.avx.rsqrt.ps.256";
|
|
|
|
}
|
|
|
|
return lp_build_intrinsic_unary(builder, intrinsic, bld->vec_type, a);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
debug_printf("%s: emulating fast rsqrt with rcp/sqrt\n", __FUNCTION__);
|
|
|
|
}
|
|
|
|
return lp_build_rcp(bld, lp_build_sqrt(bld, a));
|
|
|
|
}
|
|
|
|
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
2013-08-21 21:01:12 +01:00
|
|
|
* Generate sin(a) or cos(a) using polynomial approximation.
|
|
|
|
* TODO: it might be worth recognizing sin and cos using same source
|
|
|
|
* (i.e. d3d10 sincos opcode). Obviously doing both at the same time
|
|
|
|
* would be way cheaper than calculating (nearly) everything twice...
|
|
|
|
* Not sure it's common enough to be worth bothering however, scs
|
|
|
|
* opcode could also benefit from calculating both though.
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2013-08-21 21:01:12 +01:00
|
|
|
static LLVMValueRef
|
|
|
|
lp_build_sin_or_cos(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a,
|
|
|
|
boolean cos)
|
2009-08-11 13:09:35 +01:00
|
|
|
{
|
2010-11-30 23:07:52 +00:00
|
|
|
struct gallivm_state *gallivm = bld->gallivm;
|
2013-08-21 21:01:12 +01:00
|
|
|
LLVMBuilderRef b = gallivm->builder;
|
2010-05-24 13:41:17 +01:00
|
|
|
struct lp_type int_type = lp_int_type(bld->type);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* take the absolute value,
|
|
|
|
* x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
|
|
|
|
*/
|
|
|
|
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef inv_sig_mask = lp_build_const_int_vec(gallivm, bld->type, ~0x80000000);
|
|
|
|
LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, bld->int_vec_type, "a_v4si");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
LLVMValueRef absi = LLVMBuildAnd(b, a_v4si, inv_sig_mask, "absi");
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef x_abs = LLVMBuildBitCast(b, absi, bld->vec_type, "x_abs");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* scale by 4/Pi
|
|
|
|
* y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
|
|
|
|
*/
|
2013-08-21 21:01:12 +01:00
|
|
|
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef FOPi = lp_build_const_vec(gallivm, bld->type, 1.27323954473516);
|
2010-08-08 20:44:54 +01:00
|
|
|
LLVMValueRef scale_y = LLVMBuildFMul(b, x_abs, FOPi, "scale_y");
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2010-05-24 13:41:17 +01:00
|
|
|
/*
|
|
|
|
* store the integer part of y in mm0
|
|
|
|
* emm2 = _mm_cvttps_epi32(y);
|
|
|
|
*/
|
2013-08-21 21:01:12 +01:00
|
|
|
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef emm2_i = LLVMBuildFPToSI(b, scale_y, bld->int_vec_type, "emm2_i");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* j=(j+1) & (~1) (see the cephes sources)
|
|
|
|
* emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
|
|
|
|
*/
|
|
|
|
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef all_one = lp_build_const_int_vec(gallivm, bld->type, 1);
|
2010-05-24 13:41:17 +01:00
|
|
|
LLVMValueRef emm2_add = LLVMBuildAdd(b, emm2_i, all_one, "emm2_add");
|
|
|
|
/*
|
|
|
|
* emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
|
|
|
|
*/
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef inv_one = lp_build_const_int_vec(gallivm, bld->type, ~1);
|
2010-05-24 13:41:17 +01:00
|
|
|
LLVMValueRef emm2_and = LLVMBuildAnd(b, emm2_add, inv_one, "emm2_and");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* y = _mm_cvtepi32_ps(emm2);
|
|
|
|
*/
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef y_2 = LLVMBuildSIToFP(b, emm2_and, bld->vec_type, "y_2");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
2013-08-21 21:01:12 +01:00
|
|
|
LLVMValueRef const_2 = lp_build_const_int_vec(gallivm, bld->type, 2);
|
|
|
|
LLVMValueRef const_4 = lp_build_const_int_vec(gallivm, bld->type, 4);
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef const_29 = lp_build_const_int_vec(gallivm, bld->type, 29);
|
2013-08-21 21:01:12 +01:00
|
|
|
LLVMValueRef sign_mask = lp_build_const_int_vec(gallivm, bld->type, 0x80000000);
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
2013-08-21 21:01:12 +01:00
|
|
|
* Argument used for poly selection and sign bit determination
|
|
|
|
* is different for sin vs. cos.
|
|
|
|
*/
|
|
|
|
LLVMValueRef emm2_2 = cos ? LLVMBuildSub(b, emm2_and, const_2, "emm2_2") :
|
|
|
|
emm2_and;
|
|
|
|
|
|
|
|
LLVMValueRef sign_bit = cos ? LLVMBuildShl(b, LLVMBuildAnd(b, const_4,
|
|
|
|
LLVMBuildNot(b, emm2_2, ""), ""),
|
|
|
|
const_29, "sign_bit") :
|
|
|
|
LLVMBuildAnd(b, LLVMBuildXor(b, a_v4si,
|
|
|
|
LLVMBuildShl(b, emm2_add,
|
|
|
|
const_29, ""), ""),
|
|
|
|
sign_mask, "sign_bit");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* get the polynom selection mask
|
2010-05-24 13:41:17 +01:00
|
|
|
* there is one polynom for 0 <= x <= Pi/4
|
|
|
|
* and another one for Pi/4<x<=Pi/2
|
|
|
|
* Both branches will be computed.
|
2013-08-21 21:01:12 +01:00
|
|
|
*
|
2010-05-24 13:41:17 +01:00
|
|
|
* emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_2);
|
|
|
|
* emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
|
|
|
|
*/
|
|
|
|
|
2013-08-21 21:01:12 +01:00
|
|
|
LLVMValueRef emm2_3 = LLVMBuildAnd(b, emm2_2, const_2, "emm2_3");
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef poly_mask = lp_build_compare(gallivm,
|
2010-11-30 23:07:52 +00:00
|
|
|
int_type, PIPE_FUNC_EQUAL,
|
2012-02-17 16:22:18 +00:00
|
|
|
emm2_3, lp_build_const_int_vec(gallivm, bld->type, 0));
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _PS_CONST(minus_cephes_DP1, -0.78515625);
|
|
|
|
* _PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4);
|
|
|
|
* _PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8);
|
|
|
|
*/
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef DP1 = lp_build_const_vec(gallivm, bld->type, -0.78515625);
|
|
|
|
LLVMValueRef DP2 = lp_build_const_vec(gallivm, bld->type, -2.4187564849853515625e-4);
|
|
|
|
LLVMValueRef DP3 = lp_build_const_vec(gallivm, bld->type, -3.77489497744594108e-8);
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
2013-08-21 21:01:12 +01:00
|
|
|
* The magic pass: "Extended precision modular arithmetic"
|
|
|
|
* x = ((x - y * DP1) - y * DP2) - y * DP3;
|
|
|
|
*/
|
2016-04-04 00:05:33 +01:00
|
|
|
LLVMValueRef x_1 = lp_build_fmuladd(b, y_2, DP1, x_abs);
|
|
|
|
LLVMValueRef x_2 = lp_build_fmuladd(b, y_2, DP2, x_1);
|
|
|
|
LLVMValueRef x_3 = lp_build_fmuladd(b, y_2, DP3, x_2);
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Evaluate the first polynom (0 <= x <= Pi/4)
|
|
|
|
*
|
|
|
|
* z = _mm_mul_ps(x,x);
|
|
|
|
*/
|
2010-08-08 20:44:54 +01:00
|
|
|
LLVMValueRef z = LLVMBuildFMul(b, x_3, x_3, "z");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _PS_CONST(coscof_p0, 2.443315711809948E-005);
|
|
|
|
* _PS_CONST(coscof_p1, -1.388731625493765E-003);
|
|
|
|
* _PS_CONST(coscof_p2, 4.166664568298827E-002);
|
|
|
|
*/
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef coscof_p0 = lp_build_const_vec(gallivm, bld->type, 2.443315711809948E-005);
|
|
|
|
LLVMValueRef coscof_p1 = lp_build_const_vec(gallivm, bld->type, -1.388731625493765E-003);
|
|
|
|
LLVMValueRef coscof_p2 = lp_build_const_vec(gallivm, bld->type, 4.166664568298827E-002);
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* y = *(v4sf*)_ps_coscof_p0;
|
|
|
|
* y = _mm_mul_ps(y, z);
|
|
|
|
*/
|
2016-04-04 00:05:33 +01:00
|
|
|
LLVMValueRef y_4 = lp_build_fmuladd(b, z, coscof_p0, coscof_p1);
|
|
|
|
LLVMValueRef y_6 = lp_build_fmuladd(b, y_4, z, coscof_p2);
|
2010-08-08 20:44:54 +01:00
|
|
|
LLVMValueRef y_7 = LLVMBuildFMul(b, y_6, z, "y_7");
|
|
|
|
LLVMValueRef y_8 = LLVMBuildFMul(b, y_7, z, "y_8");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* tmp = _mm_mul_ps(z, *(v4sf*)_ps_0p5);
|
|
|
|
* y = _mm_sub_ps(y, tmp);
|
|
|
|
* y = _mm_add_ps(y, *(v4sf*)_ps_1);
|
2013-08-21 21:01:12 +01:00
|
|
|
*/
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef half = lp_build_const_vec(gallivm, bld->type, 0.5);
|
2010-08-08 20:44:54 +01:00
|
|
|
LLVMValueRef tmp = LLVMBuildFMul(b, z, half, "tmp");
|
|
|
|
LLVMValueRef y_9 = LLVMBuildFSub(b, y_8, tmp, "y_8");
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef one = lp_build_const_vec(gallivm, bld->type, 1.0);
|
2010-08-08 20:44:54 +01:00
|
|
|
LLVMValueRef y_10 = LLVMBuildFAdd(b, y_9, one, "y_9");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _PS_CONST(sincof_p0, -1.9515295891E-4);
|
|
|
|
* _PS_CONST(sincof_p1, 8.3321608736E-3);
|
|
|
|
* _PS_CONST(sincof_p2, -1.6666654611E-1);
|
|
|
|
*/
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef sincof_p0 = lp_build_const_vec(gallivm, bld->type, -1.9515295891E-4);
|
|
|
|
LLVMValueRef sincof_p1 = lp_build_const_vec(gallivm, bld->type, 8.3321608736E-3);
|
|
|
|
LLVMValueRef sincof_p2 = lp_build_const_vec(gallivm, bld->type, -1.6666654611E-1);
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Evaluate the second polynom (Pi/4 <= x <= 0)
|
|
|
|
*
|
|
|
|
* y2 = *(v4sf*)_ps_sincof_p0;
|
|
|
|
* y2 = _mm_mul_ps(y2, z);
|
|
|
|
* y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p1);
|
|
|
|
* y2 = _mm_mul_ps(y2, z);
|
|
|
|
* y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p2);
|
|
|
|
* y2 = _mm_mul_ps(y2, z);
|
|
|
|
* y2 = _mm_mul_ps(y2, x);
|
|
|
|
* y2 = _mm_add_ps(y2, x);
|
|
|
|
*/
|
|
|
|
|
2016-04-04 00:05:33 +01:00
|
|
|
LLVMValueRef y2_4 = lp_build_fmuladd(b, z, sincof_p0, sincof_p1);
|
|
|
|
LLVMValueRef y2_6 = lp_build_fmuladd(b, y2_4, z, sincof_p2);
|
2010-08-08 20:44:54 +01:00
|
|
|
LLVMValueRef y2_7 = LLVMBuildFMul(b, y2_6, z, "y2_7");
|
2016-04-04 00:05:33 +01:00
|
|
|
LLVMValueRef y2_9 = lp_build_fmuladd(b, y2_7, x_3, x_3);
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* select the correct result from the two polynoms
|
|
|
|
* xmm3 = poly_mask;
|
|
|
|
* y2 = _mm_and_ps(xmm3, y2); //, xmm3);
|
|
|
|
* y = _mm_andnot_ps(xmm3, y);
|
2013-07-17 17:13:41 +01:00
|
|
|
* y = _mm_or_ps(y,y2);
|
2010-05-24 13:41:17 +01:00
|
|
|
*/
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, bld->int_vec_type, "y2_i");
|
|
|
|
LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, bld->int_vec_type, "y_i");
|
2010-05-24 13:41:17 +01:00
|
|
|
LLVMValueRef y2_and = LLVMBuildAnd(b, y2_i, poly_mask, "y2_and");
|
2013-07-17 17:13:41 +01:00
|
|
|
LLVMValueRef poly_mask_inv = LLVMBuildNot(b, poly_mask, "poly_mask_inv");
|
2010-05-24 13:41:17 +01:00
|
|
|
LLVMValueRef y_and = LLVMBuildAnd(b, y_i, poly_mask_inv, "y_and");
|
2013-07-17 17:13:41 +01:00
|
|
|
LLVMValueRef y_combine = LLVMBuildOr(b, y_and, y2_and, "y_combine");
|
2010-05-24 13:41:17 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* update the sign
|
|
|
|
* y = _mm_xor_ps(y, sign_bit);
|
|
|
|
*/
|
2013-08-21 21:01:12 +01:00
|
|
|
LLVMValueRef y_sign = LLVMBuildXor(b, y_combine, sign_bit, "y_sign");
|
2012-02-17 16:22:18 +00:00
|
|
|
LLVMValueRef y_result = LLVMBuildBitCast(b, y_sign, bld->vec_type, "y_result");
|
2013-08-21 21:01:12 +01:00
|
|
|
|
2013-07-17 21:55:52 +01:00
|
|
|
LLVMValueRef isfinite = lp_build_isfinite(bld, a);
|
|
|
|
|
|
|
|
/* clamp output to be within [-1, 1] */
|
|
|
|
y_result = lp_build_clamp(bld, y_result,
|
|
|
|
lp_build_const_vec(bld->gallivm, bld->type, -1.f),
|
|
|
|
lp_build_const_vec(bld->gallivm, bld->type, 1.f));
|
|
|
|
/* If a is -inf, inf or NaN then return NaN */
|
|
|
|
y_result = lp_build_select(bld, isfinite, y_result,
|
|
|
|
lp_build_const_vec(bld->gallivm, bld->type, NAN));
|
2010-05-24 13:41:17 +01:00
|
|
|
return y_result;
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
2013-08-21 21:01:12 +01:00
|
|
|
* Generate sin(a)
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2009-08-11 13:09:35 +01:00
|
|
|
LLVMValueRef
|
2013-08-21 21:01:12 +01:00
|
|
|
lp_build_sin(struct lp_build_context *bld,
|
2010-05-24 13:41:17 +01:00
|
|
|
LLVMValueRef a)
|
2009-08-11 13:09:35 +01:00
|
|
|
{
|
2021-07-19 03:49:00 +01:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
|
|
|
|
if (type.width == 16) {
|
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
|
|
|
char intrinsic[32];
|
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.sin", vec_type);
|
|
|
|
LLVMValueRef args[] = { a };
|
|
|
|
return lp_build_intrinsic(builder, intrinsic, vec_type, args, 1, 0);
|
|
|
|
}
|
|
|
|
|
2013-08-21 21:01:12 +01:00
|
|
|
return lp_build_sin_or_cos(bld, a, FALSE);
|
|
|
|
}
|
2010-05-24 13:41:17 +01:00
|
|
|
|
2013-07-17 21:55:52 +01:00
|
|
|
|
2013-08-21 21:01:12 +01:00
|
|
|
/**
|
|
|
|
* Generate cos(a)
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_cos(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef a)
|
|
|
|
{
|
2021-07-19 03:49:00 +01:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
|
|
|
|
if (type.width == 16) {
|
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
|
|
|
char intrinsic[32];
|
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.cos", vec_type);
|
|
|
|
LLVMValueRef args[] = { a };
|
|
|
|
return lp_build_intrinsic(builder, intrinsic, vec_type, args, 1, 0);
|
|
|
|
}
|
|
|
|
|
2013-08-21 21:01:12 +01:00
|
|
|
return lp_build_sin_or_cos(bld, a, TRUE);
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate pow(x, y)
|
|
|
|
*/
|
2009-08-11 13:09:35 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_pow(struct lp_build_context *bld,
|
2009-08-12 10:35:50 +01:00
|
|
|
LLVMValueRef x,
|
|
|
|
LLVMValueRef y)
|
2009-08-11 13:09:35 +01:00
|
|
|
{
|
|
|
|
/* TODO: optimize the constant case */
|
2010-09-09 12:09:44 +01:00
|
|
|
if (gallivm_debug & GALLIVM_DEBUG_PERF &&
|
|
|
|
LLVMIsConstant(x) && LLVMIsConstant(y)) {
|
2009-09-24 23:08:47 +01:00
|
|
|
debug_printf("%s: inefficient/imprecise constant arithmetic\n",
|
|
|
|
__FUNCTION__);
|
2010-09-09 12:09:44 +01:00
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
|
2020-09-29 19:49:43 +01:00
|
|
|
LLVMValueRef cmp = lp_build_cmp(bld, PIPE_FUNC_EQUAL, x, lp_build_const_vec(bld->gallivm, bld->type, 0.0f));
|
2021-07-27 13:28:44 +01:00
|
|
|
LLVMValueRef res = lp_build_exp2(bld, lp_build_mul(bld, lp_build_log2_safe(bld, x), y));
|
2020-09-29 19:49:43 +01:00
|
|
|
|
|
|
|
res = lp_build_select(bld, cmp, lp_build_const_vec(bld->gallivm, bld->type, 0.0f), res);
|
|
|
|
return res;
|
2009-08-12 02:34:55 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate exp(x)
|
|
|
|
*/
|
2009-08-12 02:34:55 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_exp(struct lp_build_context *bld,
|
2009-08-12 10:35:50 +01:00
|
|
|
LLVMValueRef x)
|
2009-08-12 02:34:55 +01:00
|
|
|
{
|
2009-08-12 10:35:50 +01:00
|
|
|
/* log2(e) = 1/log(2) */
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMValueRef log2e = lp_build_const_vec(bld->gallivm, bld->type,
|
|
|
|
1.4426950408889634);
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
2011-07-19 23:58:09 +01:00
|
|
|
return lp_build_exp2(bld, lp_build_mul(bld, log2e, x));
|
2009-08-12 02:34:55 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate log(x)
|
2013-07-18 08:32:02 +01:00
|
|
|
* Behavior is undefined with infs, 0s and nans
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
2009-08-12 02:34:55 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_log(struct lp_build_context *bld,
|
2009-08-12 10:35:50 +01:00
|
|
|
LLVMValueRef x)
|
2009-08-12 02:34:55 +01:00
|
|
|
{
|
2009-08-12 10:35:50 +01:00
|
|
|
/* log(2) */
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMValueRef log2 = lp_build_const_vec(bld->gallivm, bld->type,
|
|
|
|
0.69314718055994529);
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
2011-07-19 23:58:09 +01:00
|
|
|
return lp_build_mul(bld, log2, lp_build_log2(bld, x));
|
2009-08-12 02:34:55 +01:00
|
|
|
}
|
|
|
|
|
2013-07-18 08:32:02 +01:00
|
|
|
/**
|
|
|
|
* Generate log(x) that handles edge cases (infs, 0s and nans)
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_log_safe(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
|
|
|
/* log(2) */
|
|
|
|
LLVMValueRef log2 = lp_build_const_vec(bld->gallivm, bld->type,
|
|
|
|
0.69314718055994529);
|
|
|
|
|
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
|
|
|
return lp_build_mul(bld, log2, lp_build_log2_safe(bld, x));
|
|
|
|
}
|
|
|
|
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2009-08-19 20:13:49 +01:00
|
|
|
/**
|
|
|
|
* Generate polynomial.
|
2009-12-15 14:15:52 +00:00
|
|
|
* Ex: coeffs[0] + x * coeffs[1] + x^2 * coeffs[2].
|
2009-08-19 20:13:49 +01:00
|
|
|
*/
|
gallivm: handle srgb-to-linear and linear-to-srgb conversions
srgb-to-linear is using 3rd degree polynomial for now which should be _just_
good enough. Reverse is using some rational polynomials and is quite accurate,
though not hooked into llvmpipe's blend code yet and hence unused (untested).
Using a table might also be an option (for srgb-to-linear especially).
This does not enable any new features yet because EXT_texture_srgb was already
supported via util_format fallbacks, but performance was lacking probably due
to the external function call (the table used by the util_format_srgb code may
not be all that much slower on its own).
Some performance figures (taken from modified gloss, replaced both base and
sphere texture to use GL_SRGB instead of GL_RGB, measured on 1Ghz Sandy Bridge,
the numbers aren't terribly accurate):
normal gloss, aos, 8-wide: 47 fps
normal gloss, aos, 4-wide: 48 fps
normal gloss, forced to soa, 8-wide: 48 fps
normal gloss, forced to soa, 4-wide: 47 fps
patched gloss, old code, soa, 8-wide: 21 fps
patched gloss, old code, soa, 4-wide: 24 fps
patched gloss, new code, soa, 8-wide: 41 fps
patched gloss, new code, soa, 4-wide: 38 fps
So there's a performance hit but it seems acceptable, certainly better
than using the fallback.
Note the new code only works for 4x8bit srgb formats, others (L8/L8A8) will
continue to use the old util_format fallback, because I can't be bothered
to write code for formats noone uses anyway (as decoding is done as part of
lp_build_unpack_rgba_soa which can only handle block type width of 32).
Compressed srgb formats should get their own path though eventually (it is
going to be expensive in any case, first decompress, then convert).
No piglit regressions.
v2: use lp_build_polynomial instead of ad-hoc polynomial construction, also
since keeping both linear to srgb functions for now make sure both are
compiled (since they share quite some code just integrate into the same
function).
v3: formatting fixes and bugfix in the complicated (disabled) linear-to-srgb
path.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
2013-07-13 16:31:52 +01:00
|
|
|
LLVMValueRef
|
2009-08-12 02:34:55 +01:00
|
|
|
lp_build_polynomial(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
const double *coeffs,
|
|
|
|
unsigned num_coeffs)
|
|
|
|
{
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2012-04-05 20:32:51 +01:00
|
|
|
LLVMValueRef even = NULL, odd = NULL;
|
|
|
|
LLVMValueRef x2;
|
2009-08-12 02:34:55 +01:00
|
|
|
unsigned i;
|
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
2009-08-12 10:35:50 +01:00
|
|
|
/* TODO: optimize the constant case */
|
2010-09-09 12:09:44 +01:00
|
|
|
if (gallivm_debug & GALLIVM_DEBUG_PERF &&
|
|
|
|
LLVMIsConstant(x)) {
|
2009-09-24 23:08:47 +01:00
|
|
|
debug_printf("%s: inefficient/imprecise constant arithmetic\n",
|
|
|
|
__FUNCTION__);
|
2010-09-09 12:09:44 +01:00
|
|
|
}
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2012-04-05 20:32:51 +01:00
|
|
|
/*
|
|
|
|
* Calculate odd and even terms seperately to decrease data dependency
|
|
|
|
* Ex:
|
|
|
|
* c[0] + x^2 * c[2] + x^4 * c[4] ...
|
|
|
|
* + x * (c[1] + x^2 * c[3] + x^4 * c[5]) ...
|
|
|
|
*/
|
|
|
|
x2 = lp_build_mul(bld, x, x);
|
|
|
|
|
2009-08-12 02:34:55 +01:00
|
|
|
for (i = num_coeffs; i--; ) {
|
2010-03-09 16:37:50 +00:00
|
|
|
LLVMValueRef coeff;
|
|
|
|
|
2010-11-30 23:07:52 +00:00
|
|
|
coeff = lp_build_const_vec(bld->gallivm, type, coeffs[i]);
|
2010-03-09 16:37:50 +00:00
|
|
|
|
2012-04-05 20:32:51 +01:00
|
|
|
if (i % 2 == 0) {
|
|
|
|
if (even)
|
2016-04-04 00:05:33 +01:00
|
|
|
even = lp_build_mad(bld, x2, even, coeff);
|
2012-04-05 20:32:51 +01:00
|
|
|
else
|
|
|
|
even = coeff;
|
|
|
|
} else {
|
|
|
|
if (odd)
|
2016-04-04 00:05:33 +01:00
|
|
|
odd = lp_build_mad(bld, x2, odd, coeff);
|
2012-04-05 20:32:51 +01:00
|
|
|
else
|
|
|
|
odd = coeff;
|
|
|
|
}
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
2012-04-05 20:32:51 +01:00
|
|
|
if (odd)
|
2016-04-04 00:05:33 +01:00
|
|
|
return lp_build_mad(bld, odd, x, even);
|
2012-04-05 20:32:51 +01:00
|
|
|
else if (even)
|
|
|
|
return even;
|
2009-08-12 02:34:55 +01:00
|
|
|
else
|
|
|
|
return bld->undef;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-08-12 10:35:50 +01:00
|
|
|
/**
|
2010-05-08 21:42:55 +01:00
|
|
|
* Minimax polynomial fit of 2**x, in range [0, 1[
|
2009-08-12 10:35:50 +01:00
|
|
|
*/
|
|
|
|
const double lp_build_exp2_polynomial[] = {
|
|
|
|
#if EXP_POLY_DEGREE == 5
|
2013-07-18 07:38:54 +01:00
|
|
|
1.000000000000000000000, /*XXX: was 0.999999925063526176901, recompute others */
|
2011-07-20 22:53:08 +01:00
|
|
|
0.693153073200168932794,
|
|
|
|
0.240153617044375388211,
|
|
|
|
0.0558263180532956664775,
|
|
|
|
0.00898934009049466391101,
|
|
|
|
0.00187757667519147912699
|
2009-08-12 10:35:50 +01:00
|
|
|
#elif EXP_POLY_DEGREE == 4
|
2011-07-20 22:53:08 +01:00
|
|
|
1.00000259337069434683,
|
|
|
|
0.693003834469974940458,
|
|
|
|
0.24144275689150793076,
|
|
|
|
0.0520114606103070150235,
|
|
|
|
0.0135341679161270268764
|
2009-08-12 10:35:50 +01:00
|
|
|
#elif EXP_POLY_DEGREE == 3
|
2010-05-08 21:42:55 +01:00
|
|
|
0.999925218562710312959,
|
|
|
|
0.695833540494823811697,
|
|
|
|
0.226067155427249155588,
|
|
|
|
0.0780245226406372992967
|
2009-08-12 10:35:50 +01:00
|
|
|
#elif EXP_POLY_DEGREE == 2
|
2010-05-08 21:42:55 +01:00
|
|
|
1.00172476321474503578,
|
|
|
|
0.657636275736077639316,
|
|
|
|
0.33718943461968720704
|
2009-08-12 10:35:50 +01:00
|
|
|
#else
|
|
|
|
#error
|
|
|
|
#endif
|
|
|
|
};
|
|
|
|
|
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_exp2(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
2009-08-12 02:34:55 +01:00
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
2009-08-16 21:07:15 +01:00
|
|
|
LLVMValueRef ipart = NULL;
|
|
|
|
LLVMValueRef fpart = NULL;
|
|
|
|
LLVMValueRef expipart = NULL;
|
|
|
|
LLVMValueRef expfpart = NULL;
|
|
|
|
LLVMValueRef res = NULL;
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2021-07-22 06:36:59 +01:00
|
|
|
if (type.floating && type.width == 16) {
|
|
|
|
char intrinsic[32];
|
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.exp2", vec_type);
|
|
|
|
LLVMValueRef args[] = { x };
|
|
|
|
return lp_build_intrinsic(builder, intrinsic, vec_type, args, 1, 0);
|
|
|
|
}
|
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
/* TODO: optimize the constant case */
|
|
|
|
if (gallivm_debug & GALLIVM_DEBUG_PERF &&
|
|
|
|
LLVMIsConstant(x)) {
|
|
|
|
debug_printf("%s: inefficient/imprecise constant arithmetic\n",
|
|
|
|
__FUNCTION__);
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(type.floating && type.width == 32);
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
/* We want to preserve NaN and make sure than for exp2 if x > 128,
|
|
|
|
* the result is INF and if it's smaller than -126.9 the result is 0 */
|
|
|
|
x = lp_build_min_ext(bld, lp_build_const_vec(bld->gallivm, type, 128.0), x,
|
2014-08-29 04:12:00 +01:00
|
|
|
GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN);
|
|
|
|
x = lp_build_max_ext(bld, lp_build_const_vec(bld->gallivm, type, -126.99999),
|
|
|
|
x, GALLIVM_NAN_RETURN_NAN_FIRST_NONNAN);
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
/* ipart = floor(x) */
|
|
|
|
/* fpart = x - ipart */
|
|
|
|
lp_build_ifloor_fract(bld, x, &ipart, &fpart);
|
2009-08-12 12:42:06 +01:00
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
/* expipart = (float) (1 << ipart) */
|
|
|
|
expipart = LLVMBuildAdd(builder, ipart,
|
|
|
|
lp_build_const_int_vec(bld->gallivm, type, 127), "");
|
|
|
|
expipart = LLVMBuildShl(builder, expipart,
|
|
|
|
lp_build_const_int_vec(bld->gallivm, type, 23), "");
|
|
|
|
expipart = LLVMBuildBitCast(builder, expipart, vec_type, "");
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
expfpart = lp_build_polynomial(bld, fpart, lp_build_exp2_polynomial,
|
2016-04-25 22:56:08 +01:00
|
|
|
ARRAY_SIZE(lp_build_exp2_polynomial));
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
res = LLVMBuildFMul(builder, expipart, expfpart, "");
|
2009-08-12 12:42:06 +01:00
|
|
|
|
|
|
|
return res;
|
2009-08-11 13:09:35 +01:00
|
|
|
}
|
|
|
|
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2013-09-11 12:04:29 +01:00
|
|
|
|
2010-10-09 12:11:20 +01:00
|
|
|
/**
|
|
|
|
* Extract the exponent of a IEEE-754 floating point value.
|
|
|
|
*
|
|
|
|
* Optionally apply an integer bias.
|
|
|
|
*
|
|
|
|
* Result is an integer value with
|
|
|
|
*
|
|
|
|
* ifloor(log2(x)) + bias
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_extract_exponent(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
int bias)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-10-09 12:11:20 +01:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
unsigned mantissa = lp_mantissa(type);
|
|
|
|
LLVMValueRef res;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
|
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
x = LLVMBuildBitCast(builder, x, bld->int_vec_type, "");
|
2010-10-09 12:11:20 +01:00
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildLShr(builder, x,
|
2010-11-30 23:07:52 +00:00
|
|
|
lp_build_const_int_vec(bld->gallivm, type, mantissa), "");
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildAnd(builder, res,
|
2010-11-30 23:07:52 +00:00
|
|
|
lp_build_const_int_vec(bld->gallivm, type, 255), "");
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildSub(builder, res,
|
2010-11-30 23:07:52 +00:00
|
|
|
lp_build_const_int_vec(bld->gallivm, type, 127 - bias), "");
|
2010-10-09 12:11:20 +01:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Extract the mantissa of the a floating.
|
|
|
|
*
|
|
|
|
* Result is a floating point value with
|
|
|
|
*
|
|
|
|
* x / floor(log2(x))
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_extract_mantissa(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-10-09 12:11:20 +01:00
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
unsigned mantissa = lp_mantissa(type);
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMValueRef mantmask = lp_build_const_int_vec(bld->gallivm, type,
|
|
|
|
(1ULL << mantissa) - 1);
|
2010-10-09 12:11:20 +01:00
|
|
|
LLVMValueRef one = LLVMConstBitCast(bld->one, bld->int_vec_type);
|
|
|
|
LLVMValueRef res;
|
|
|
|
|
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
x = LLVMBuildBitCast(builder, x, bld->int_vec_type, "");
|
2010-10-09 12:11:20 +01:00
|
|
|
|
|
|
|
/* res = x / 2**ipart */
|
2010-12-03 01:11:13 +00:00
|
|
|
res = LLVMBuildAnd(builder, x, mantmask, "");
|
|
|
|
res = LLVMBuildOr(builder, res, one, "");
|
|
|
|
res = LLVMBuildBitCast(builder, res, bld->vec_type, "");
|
2010-10-09 12:11:20 +01:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-08-12 10:35:50 +01:00
|
|
|
/**
|
2012-04-05 20:34:11 +01:00
|
|
|
* Minimax polynomial fit of log2((1.0 + sqrt(x))/(1.0 - sqrt(x)))/sqrt(x) ,for x in range of [0, 1/9[
|
2009-08-12 10:35:50 +01:00
|
|
|
* These coefficients can be generate with
|
|
|
|
* http://www.boost.org/doc/libs/1_36_0/libs/math/doc/sf_and_dist/html/math_toolkit/toolkit/internals2/minimax.html
|
|
|
|
*/
|
|
|
|
const double lp_build_log2_polynomial[] = {
|
2012-04-05 20:34:11 +01:00
|
|
|
#if LOG_POLY_DEGREE == 5
|
|
|
|
2.88539008148777786488L,
|
|
|
|
0.961796878841293367824L,
|
|
|
|
0.577058946784739859012L,
|
|
|
|
0.412914355135828735411L,
|
|
|
|
0.308591899232910175289L,
|
|
|
|
0.352376952300281371868L,
|
2009-08-12 10:35:50 +01:00
|
|
|
#elif LOG_POLY_DEGREE == 4
|
2012-04-05 20:34:11 +01:00
|
|
|
2.88539009343309178325L,
|
|
|
|
0.961791550404184197881L,
|
|
|
|
0.577440339438736392009L,
|
|
|
|
0.403343858251329912514L,
|
|
|
|
0.406718052498846252698L,
|
2009-08-12 10:35:50 +01:00
|
|
|
#elif LOG_POLY_DEGREE == 3
|
2012-04-05 20:34:11 +01:00
|
|
|
2.88538959748872753838L,
|
|
|
|
0.961932915889597772928L,
|
|
|
|
0.571118517972136195241L,
|
|
|
|
0.493997535084709500285L,
|
2009-08-12 10:35:50 +01:00
|
|
|
#else
|
|
|
|
#error
|
|
|
|
#endif
|
|
|
|
};
|
|
|
|
|
2009-08-12 02:34:55 +01:00
|
|
|
/**
|
|
|
|
* See http://www.devmaster.net/forums/showthread.php?p=43580
|
2012-04-05 20:34:11 +01:00
|
|
|
* http://en.wikipedia.org/wiki/Logarithm#Calculation
|
|
|
|
* http://www.nezumi.demon.co.uk/consult/logx.htm
|
2013-07-16 20:57:11 +01:00
|
|
|
*
|
|
|
|
* If handle_edge_cases is true the function will perform computations
|
|
|
|
* to match the required D3D10+ behavior for each of the edge cases.
|
|
|
|
* That means that if input is:
|
|
|
|
* - less than zero (to and including -inf) then NaN will be returned
|
|
|
|
* - equal to zero (-denorm, -0, +0 or +denorm), then -inf will be returned
|
|
|
|
* - +infinity, then +infinity will be returned
|
|
|
|
* - NaN, then NaN will be returned
|
|
|
|
*
|
|
|
|
* Those checks are fairly expensive so if you don't need them make sure
|
|
|
|
* handle_edge_cases is false.
|
2009-08-12 02:34:55 +01:00
|
|
|
*/
|
2009-08-12 12:42:06 +01:00
|
|
|
void
|
|
|
|
lp_build_log2_approx(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
LLVMValueRef *p_exp,
|
|
|
|
LLVMValueRef *p_floor_log2,
|
2013-07-16 20:57:11 +01:00
|
|
|
LLVMValueRef *p_log2,
|
|
|
|
boolean handle_edge_cases)
|
2009-08-12 02:34:55 +01:00
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2009-09-14 11:05:06 +01:00
|
|
|
const struct lp_type type = bld->type;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMTypeRef vec_type = lp_build_vec_type(bld->gallivm, type);
|
|
|
|
LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, type);
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMValueRef expmask = lp_build_const_int_vec(bld->gallivm, type, 0x7f800000);
|
|
|
|
LLVMValueRef mantmask = lp_build_const_int_vec(bld->gallivm, type, 0x007fffff);
|
2009-08-12 02:34:55 +01:00
|
|
|
LLVMValueRef one = LLVMConstBitCast(bld->one, int_vec_type);
|
|
|
|
|
2009-08-16 21:07:15 +01:00
|
|
|
LLVMValueRef i = NULL;
|
2012-04-05 20:34:11 +01:00
|
|
|
LLVMValueRef y = NULL;
|
|
|
|
LLVMValueRef z = NULL;
|
2009-08-16 21:07:15 +01:00
|
|
|
LLVMValueRef exp = NULL;
|
|
|
|
LLVMValueRef mant = NULL;
|
|
|
|
LLVMValueRef logexp = NULL;
|
2016-04-04 00:05:33 +01:00
|
|
|
LLVMValueRef p_z = NULL;
|
2009-08-16 21:07:15 +01:00
|
|
|
LLVMValueRef res = NULL;
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2021-07-22 06:36:59 +01:00
|
|
|
if (bld->type.width == 16) {
|
|
|
|
char intrinsic[32];
|
|
|
|
lp_format_intrinsic(intrinsic, sizeof intrinsic, "llvm.log2", bld->vec_type);
|
|
|
|
LLVMValueRef args[] = { x };
|
|
|
|
if (p_log2)
|
|
|
|
*p_log2 = lp_build_intrinsic(builder, intrinsic, bld->vec_type, args, 1, 0);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2010-08-09 17:30:33 +01:00
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
2009-08-12 12:42:06 +01:00
|
|
|
if(p_exp || p_floor_log2 || p_log2) {
|
|
|
|
/* TODO: optimize the constant case */
|
2010-09-09 12:09:44 +01:00
|
|
|
if (gallivm_debug & GALLIVM_DEBUG_PERF &&
|
|
|
|
LLVMIsConstant(x)) {
|
2009-09-24 23:08:47 +01:00
|
|
|
debug_printf("%s: inefficient/imprecise constant arithmetic\n",
|
|
|
|
__FUNCTION__);
|
2010-09-09 12:09:44 +01:00
|
|
|
}
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2009-08-12 12:42:06 +01:00
|
|
|
assert(type.floating && type.width == 32);
|
2009-08-12 10:35:50 +01:00
|
|
|
|
2011-07-20 22:53:59 +01:00
|
|
|
/*
|
|
|
|
* We don't explicitly handle denormalized numbers. They will yield a
|
|
|
|
* result in the neighbourhood of -127, which appears to be adequate
|
|
|
|
* enough.
|
|
|
|
*/
|
|
|
|
|
2010-12-03 01:11:13 +00:00
|
|
|
i = LLVMBuildBitCast(builder, x, int_vec_type, "");
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2009-08-12 12:42:06 +01:00
|
|
|
/* exp = (float) exponent(x) */
|
2010-12-03 01:11:13 +00:00
|
|
|
exp = LLVMBuildAnd(builder, i, expmask, "");
|
2009-08-12 12:42:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if(p_floor_log2 || p_log2) {
|
2010-12-03 01:11:13 +00:00
|
|
|
logexp = LLVMBuildLShr(builder, exp, lp_build_const_int_vec(bld->gallivm, type, 23), "");
|
|
|
|
logexp = LLVMBuildSub(builder, logexp, lp_build_const_int_vec(bld->gallivm, type, 127), "");
|
|
|
|
logexp = LLVMBuildSIToFP(builder, logexp, vec_type, "");
|
2009-08-12 12:42:06 +01:00
|
|
|
}
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (p_log2) {
|
2012-04-05 20:34:11 +01:00
|
|
|
/* mant = 1 + (float) mantissa(x) */
|
2010-12-03 01:11:13 +00:00
|
|
|
mant = LLVMBuildAnd(builder, i, mantmask, "");
|
|
|
|
mant = LLVMBuildOr(builder, mant, one, "");
|
|
|
|
mant = LLVMBuildBitCast(builder, mant, vec_type, "");
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2012-04-05 20:34:11 +01:00
|
|
|
/* y = (mant - 1) / (mant + 1) */
|
|
|
|
y = lp_build_div(bld,
|
|
|
|
lp_build_sub(bld, mant, bld->one),
|
|
|
|
lp_build_add(bld, mant, bld->one)
|
|
|
|
);
|
|
|
|
|
|
|
|
/* z = y^2 */
|
|
|
|
z = lp_build_mul(bld, y, y);
|
|
|
|
|
|
|
|
/* compute P(z) */
|
2016-04-04 00:05:33 +01:00
|
|
|
p_z = lp_build_polynomial(bld, z, lp_build_log2_polynomial,
|
|
|
|
ARRAY_SIZE(lp_build_log2_polynomial));
|
2009-08-12 12:42:06 +01:00
|
|
|
|
2016-04-04 00:05:33 +01:00
|
|
|
/* y * P(z) + logexp */
|
|
|
|
res = lp_build_mad(bld, y, p_z, logexp);
|
2013-07-16 20:57:11 +01:00
|
|
|
|
|
|
|
if (type.floating && handle_edge_cases) {
|
|
|
|
LLVMValueRef negmask, infmask, zmask;
|
|
|
|
negmask = lp_build_cmp(bld, PIPE_FUNC_LESS, x,
|
|
|
|
lp_build_const_vec(bld->gallivm, type, 0.0f));
|
|
|
|
zmask = lp_build_cmp(bld, PIPE_FUNC_EQUAL, x,
|
|
|
|
lp_build_const_vec(bld->gallivm, type, 0.0f));
|
|
|
|
infmask = lp_build_cmp(bld, PIPE_FUNC_GEQUAL, x,
|
|
|
|
lp_build_const_vec(bld->gallivm, type, INFINITY));
|
|
|
|
|
|
|
|
/* If x is qual to inf make sure we return inf */
|
|
|
|
res = lp_build_select(bld, infmask,
|
|
|
|
lp_build_const_vec(bld->gallivm, type, INFINITY),
|
|
|
|
res);
|
|
|
|
/* If x is qual to 0, return -inf */
|
|
|
|
res = lp_build_select(bld, zmask,
|
|
|
|
lp_build_const_vec(bld->gallivm, type, -INFINITY),
|
|
|
|
res);
|
|
|
|
/* If x is nan or less than 0, return nan */
|
|
|
|
res = lp_build_select(bld, negmask,
|
|
|
|
lp_build_const_vec(bld->gallivm, type, NAN),
|
|
|
|
res);
|
|
|
|
}
|
2009-08-12 12:42:06 +01:00
|
|
|
}
|
2009-08-12 02:34:55 +01:00
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (p_exp) {
|
2010-12-03 01:11:13 +00:00
|
|
|
exp = LLVMBuildBitCast(builder, exp, vec_type, "");
|
2009-08-12 12:42:06 +01:00
|
|
|
*p_exp = exp;
|
2010-04-22 18:00:48 +01:00
|
|
|
}
|
2009-08-12 12:42:06 +01:00
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (p_floor_log2)
|
2009-08-12 12:42:06 +01:00
|
|
|
*p_floor_log2 = logexp;
|
|
|
|
|
2015-12-04 06:12:30 +00:00
|
|
|
if (p_log2)
|
2009-08-12 12:42:06 +01:00
|
|
|
*p_log2 = res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2013-07-16 20:57:11 +01:00
|
|
|
/*
|
|
|
|
* log2 implementation which doesn't have special code to
|
|
|
|
* handle edge cases (-inf, 0, inf, NaN). It's faster but
|
|
|
|
* the results for those cases are undefined.
|
|
|
|
*/
|
2009-08-12 12:42:06 +01:00
|
|
|
LLVMValueRef
|
|
|
|
lp_build_log2(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
|
|
|
LLVMValueRef res;
|
2013-07-16 20:57:11 +01:00
|
|
|
lp_build_log2_approx(bld, x, NULL, NULL, &res, FALSE);
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Version of log2 which handles all edge cases.
|
|
|
|
* Look at documentation of lp_build_log2_approx for
|
|
|
|
* description of the behavior for each of the edge cases.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_log2_safe(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
|
|
|
LLVMValueRef res;
|
|
|
|
lp_build_log2_approx(bld, x, NULL, NULL, &res, TRUE);
|
2009-08-12 12:42:06 +01:00
|
|
|
return res;
|
2009-08-12 02:34:55 +01:00
|
|
|
}
|
2010-10-06 14:06:14 +01:00
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Faster (and less accurate) log2.
|
|
|
|
*
|
2010-10-09 12:11:20 +01:00
|
|
|
* log2(x) = floor(log2(x)) - 1 + x / 2**floor(log2(x))
|
|
|
|
*
|
|
|
|
* Piece-wise linear approximation, with exact results when x is a
|
|
|
|
* power of two.
|
2010-10-06 14:06:14 +01:00
|
|
|
*
|
|
|
|
* See http://www.flipcode.com/archives/Fast_log_Function.shtml
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_fast_log2(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-10-06 14:06:14 +01:00
|
|
|
LLVMValueRef ipart;
|
|
|
|
LLVMValueRef fpart;
|
|
|
|
|
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
2010-10-09 12:11:20 +01:00
|
|
|
assert(bld->type.floating);
|
2010-10-06 14:06:14 +01:00
|
|
|
|
|
|
|
/* ipart = floor(log2(x)) - 1 */
|
2010-10-09 12:11:20 +01:00
|
|
|
ipart = lp_build_extract_exponent(bld, x, -1);
|
2010-12-03 01:11:13 +00:00
|
|
|
ipart = LLVMBuildSIToFP(builder, ipart, bld->vec_type, "");
|
2010-10-06 14:06:14 +01:00
|
|
|
|
2010-10-09 12:11:20 +01:00
|
|
|
/* fpart = x / 2**ipart */
|
|
|
|
fpart = lp_build_extract_mantissa(bld, x);
|
2010-10-06 14:06:14 +01:00
|
|
|
|
2010-10-09 12:11:20 +01:00
|
|
|
/* ipart + fpart */
|
2010-12-03 01:11:13 +00:00
|
|
|
return LLVMBuildFAdd(builder, ipart, fpart, "");
|
2010-10-06 14:06:14 +01:00
|
|
|
}
|
2010-10-06 17:44:05 +01:00
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Fast implementation of iround(log2(x)).
|
|
|
|
*
|
|
|
|
* Not an approximation -- it should give accurate results all the time.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_ilog2(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
2010-12-03 01:11:13 +00:00
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
2010-11-30 23:07:52 +00:00
|
|
|
LLVMValueRef sqrt2 = lp_build_const_vec(bld->gallivm, bld->type, M_SQRT2);
|
2010-10-06 17:44:05 +01:00
|
|
|
LLVMValueRef ipart;
|
|
|
|
|
2010-10-09 12:11:20 +01:00
|
|
|
assert(bld->type.floating);
|
2010-10-06 17:44:05 +01:00
|
|
|
|
2010-10-09 12:11:20 +01:00
|
|
|
assert(lp_check_value(bld->type, x));
|
2010-10-06 17:44:05 +01:00
|
|
|
|
|
|
|
/* x * 2^(0.5) i.e., add 0.5 to the log2(x) */
|
2010-12-03 01:11:13 +00:00
|
|
|
x = LLVMBuildFMul(builder, x, sqrt2, "");
|
2010-10-06 17:44:05 +01:00
|
|
|
|
|
|
|
/* ipart = floor(log2(x) + 0.5) */
|
2010-10-09 12:11:20 +01:00
|
|
|
ipart = lp_build_extract_exponent(bld, x, 0);
|
2010-10-06 17:44:05 +01:00
|
|
|
|
|
|
|
return ipart;
|
|
|
|
}
|
2012-02-17 19:02:23 +00:00
|
|
|
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_mod(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x,
|
|
|
|
LLVMValueRef y)
|
|
|
|
{
|
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
LLVMValueRef res;
|
|
|
|
const struct lp_type type = bld->type;
|
|
|
|
|
|
|
|
assert(lp_check_value(type, x));
|
|
|
|
assert(lp_check_value(type, y));
|
|
|
|
|
2012-02-28 10:43:14 +00:00
|
|
|
if (type.floating)
|
|
|
|
res = LLVMBuildFRem(builder, x, y, "");
|
2012-03-05 18:09:56 +00:00
|
|
|
else if (type.sign)
|
2012-02-17 19:02:23 +00:00
|
|
|
res = LLVMBuildSRem(builder, x, y, "");
|
|
|
|
else
|
|
|
|
res = LLVMBuildURem(builder, x, y, "");
|
|
|
|
return res;
|
|
|
|
}
|
2013-07-16 18:06:24 +01:00
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* For floating inputs it creates and returns a mask
|
|
|
|
* which is all 1's for channels which are NaN.
|
|
|
|
* Channels inside x which are not NaN will be 0.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_isnan(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
|
|
|
LLVMValueRef mask;
|
|
|
|
LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, bld->type);
|
|
|
|
|
|
|
|
assert(bld->type.floating);
|
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
|
|
|
|
mask = LLVMBuildFCmp(bld->gallivm->builder, LLVMRealOEQ, x, x,
|
|
|
|
"isnotnan");
|
|
|
|
mask = LLVMBuildNot(bld->gallivm->builder, mask, "");
|
|
|
|
mask = LLVMBuildSExt(bld->gallivm->builder, mask, int_vec_type, "isnan");
|
|
|
|
return mask;
|
|
|
|
}
|
2013-07-17 21:55:52 +01:00
|
|
|
|
|
|
|
/* Returns all 1's for floating point numbers that are
|
|
|
|
* finite numbers and returns all zeros for -inf,
|
|
|
|
* inf and nan's */
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_isfinite(struct lp_build_context *bld,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
|
|
|
LLVMBuilderRef builder = bld->gallivm->builder;
|
|
|
|
LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->gallivm, bld->type);
|
|
|
|
struct lp_type int_type = lp_int_type(bld->type);
|
|
|
|
LLVMValueRef intx = LLVMBuildBitCast(builder, x, int_vec_type, "");
|
|
|
|
LLVMValueRef infornan32 = lp_build_const_int_vec(bld->gallivm, bld->type,
|
|
|
|
0x7f800000);
|
|
|
|
|
|
|
|
if (!bld->type.floating) {
|
|
|
|
return lp_build_const_int_vec(bld->gallivm, bld->type, 0);
|
|
|
|
}
|
|
|
|
assert(bld->type.floating);
|
|
|
|
assert(lp_check_value(bld->type, x));
|
|
|
|
assert(bld->type.width == 32);
|
|
|
|
|
|
|
|
intx = LLVMBuildAnd(builder, intx, infornan32, "");
|
|
|
|
return lp_build_compare(bld->gallivm, int_type, PIPE_FUNC_NOTEQUAL,
|
|
|
|
intx, infornan32);
|
|
|
|
}
|
2013-08-15 18:10:22 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Returns true if the number is nan or inf and false otherwise.
|
|
|
|
* The input has to be a floating point vector.
|
|
|
|
*/
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_is_inf_or_nan(struct gallivm_state *gallivm,
|
|
|
|
const struct lp_type type,
|
|
|
|
LLVMValueRef x)
|
|
|
|
{
|
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
|
|
|
struct lp_type int_type = lp_int_type(type);
|
|
|
|
LLVMValueRef const0 = lp_build_const_int_vec(gallivm, int_type,
|
|
|
|
0x7f800000);
|
|
|
|
LLVMValueRef ret;
|
|
|
|
|
|
|
|
assert(type.floating);
|
|
|
|
|
|
|
|
ret = LLVMBuildBitCast(builder, x, lp_build_vec_type(gallivm, int_type), "");
|
|
|
|
ret = LLVMBuildAnd(builder, ret, const0, "");
|
|
|
|
ret = lp_build_compare(gallivm, int_type, PIPE_FUNC_EQUAL,
|
|
|
|
ret, const0);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-12-06 06:28:25 +00:00
|
|
|
|
|
|
|
LLVMValueRef
|
|
|
|
lp_build_fpstate_get(struct gallivm_state *gallivm)
|
|
|
|
{
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse) {
|
2013-12-06 06:28:25 +00:00
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
|
|
|
LLVMValueRef mxcsr_ptr = lp_build_alloca(
|
|
|
|
gallivm,
|
|
|
|
LLVMInt32TypeInContext(gallivm->context),
|
|
|
|
"mxcsr_ptr");
|
2013-12-13 20:20:05 +00:00
|
|
|
LLVMValueRef mxcsr_ptr8 = LLVMBuildPointerCast(builder, mxcsr_ptr,
|
|
|
|
LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0), "");
|
2013-12-06 06:28:25 +00:00
|
|
|
lp_build_intrinsic(builder,
|
|
|
|
"llvm.x86.sse.stmxcsr",
|
|
|
|
LLVMVoidTypeInContext(gallivm->context),
|
2015-07-25 10:17:48 +01:00
|
|
|
&mxcsr_ptr8, 1, 0);
|
2013-12-06 06:28:25 +00:00
|
|
|
return mxcsr_ptr;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
lp_build_fpstate_set_denorms_zero(struct gallivm_state *gallivm,
|
|
|
|
boolean zero)
|
|
|
|
{
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse) {
|
2013-12-06 06:28:25 +00:00
|
|
|
/* turn on DAZ (64) | FTZ (32768) = 32832 if available */
|
|
|
|
int daz_ftz = _MM_FLUSH_ZERO_MASK;
|
|
|
|
|
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
|
|
|
LLVMValueRef mxcsr_ptr = lp_build_fpstate_get(gallivm);
|
|
|
|
LLVMValueRef mxcsr =
|
|
|
|
LLVMBuildLoad(builder, mxcsr_ptr, "mxcsr");
|
|
|
|
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_daz) {
|
2013-12-06 06:28:25 +00:00
|
|
|
/* Enable denormals are zero mode */
|
|
|
|
daz_ftz |= _MM_DENORMALS_ZERO_MASK;
|
|
|
|
}
|
|
|
|
if (zero) {
|
|
|
|
mxcsr = LLVMBuildOr(builder, mxcsr,
|
|
|
|
LLVMConstInt(LLVMTypeOf(mxcsr), daz_ftz, 0), "");
|
|
|
|
} else {
|
|
|
|
mxcsr = LLVMBuildAnd(builder, mxcsr,
|
|
|
|
LLVMConstInt(LLVMTypeOf(mxcsr), ~daz_ftz, 0), "");
|
|
|
|
}
|
|
|
|
|
|
|
|
LLVMBuildStore(builder, mxcsr, mxcsr_ptr);
|
|
|
|
lp_build_fpstate_set(gallivm, mxcsr_ptr);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
lp_build_fpstate_set(struct gallivm_state *gallivm,
|
|
|
|
LLVMValueRef mxcsr_ptr)
|
|
|
|
{
|
2021-02-25 18:04:50 +00:00
|
|
|
if (util_get_cpu_caps()->has_sse) {
|
2013-12-13 20:20:05 +00:00
|
|
|
LLVMBuilderRef builder = gallivm->builder;
|
|
|
|
mxcsr_ptr = LLVMBuildPointerCast(builder, mxcsr_ptr,
|
|
|
|
LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0), "");
|
|
|
|
lp_build_intrinsic(builder,
|
2013-12-06 06:28:25 +00:00
|
|
|
"llvm.x86.sse.ldmxcsr",
|
|
|
|
LLVMVoidTypeInContext(gallivm->context),
|
2015-07-25 10:17:48 +01:00
|
|
|
&mxcsr_ptr, 1, 0);
|
2013-12-06 06:28:25 +00:00
|
|
|
}
|
|
|
|
}
|