We want to replace ilo_zs_surface with ilo_state_zs. One noteworthy
difference is that ilo_state_zs always aligns level 0 to 8x4 when HiZ is
enabled. HiZ will not be enabled for 1D surfaces as a result.
It needs be set to R/W only when using certain messages via DP render cache.
Since we only use RT wrties with the render cache, we never need to set it.
Silence the warnings about the future incompatibility with automake 2.0
Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Rather than forcing everyone to provide their own definition of the symbol
provide a common (dummy) one.
This helps us resolve the build of the standalone pipe-drivers (amongst
others), which are missing the symbol.
Cc: Rob Clark <robclark@freedesktop.org>
Cc: "10.6" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Just implement it in terms of util_resource_copy_region(). Both the
original code and util_resource_copy_region() boil down to mapping,
calling util_copy_box() and unmapping.
No piglit regressions. This will also help to implement GL_ARB_copy_image.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
When we import a dma-buf fd from another driver the kernel
gives us the right info, and this trashes it.
Convert the kernel bo flags into the domain flags.
This helps getting reverse prime and glamor working.
Cc: mesa-stable@lists.freedesktop.org
Acked-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
On Lollipop, apparently stlport is gone and libcxx must be used instead.
We still support stlport when building on earlier android releases.
Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Eric Anholt <eric@anholt.net>
Based on the nice work of Paulo Sergio Travaglia <pstglia@gmail.com>.
The main modifications are:
- Include paths for LLVM header files and shared/static libraries
- Set C++ flag "c++11" to avoid compiling errors on LLVM header files
- Set defines for LLVM
- Add GALLIVM source files
- Changes path of libelf library for lollipop
Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Acked-by: Eric Anholt <eric@anholt.net>
Based on the corresponding SI support. Same as that, this is currently
only enabled for one-dimensional buffer copies due to issues with
multi-dimensional SDMA copies.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Softpipe, llvmpipe, r300g, and radeonsi pass tests. Other drivers need testing.
Freedreno and nv30 are definitely broken. Other drivers seem to be alright.
Pretty trivial, fixes the issue that we're expected to be able to blit
stencil surfaces (as the blit just relies on util blitter code which needs
stencil export to do it).
2 piglits skip->pass, 11 fail->pass
v2: prettify, keep different stencil ref value handling out of depth/stencil
test itself.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
We were returning the most recently freed BO, without checking if it
was idle yet. This meant that we generally stalled immediately on the
previous frame when generating a new one. Instead, allocate new BOs
when the *oldest* BO is still busy, so that the cache scales with how
much is needed to keep some frames outstanding, as originally
intended.
Note that if you don't have some throttling happening, this means that
you can accidentally run the system out of memory. The kernel is now
applying some throttling on all execs, to hopefully avoid this.
Roland pointed out my previous attempt was lacking, so I enhanced the
texwrap piglit test, and tested them. This fixes the offset calculations
in a number of areas by adding the offset first, it also fixes the fastpaths,
which I forgot to address in the previous commit.
v2: try and avoid divides in most paths, the repeat mirror path
really was ugly no matter which way I went, so I left it having
the divide.
Also fix the gather lod calculation bug.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The idea I had when I wrote the original shadow code was that you'd see a
set_index_buffer to the IB, then a bunch of draws out of it. What's
actually happening in openarena is that set_index_buffer occurs at every
draw, so we end up making a new shadow BO every time, and converting more
of the BO than is actually used in the draw.
While I could maybe come up with a better caching scheme, for now just
do the simple thing that doesn't result in a new shadow IB allocation
per draw.
Improves performance of isosurf in drawelements mode by 58.7967% +/-
3.86152% (n=8).
We'd sometimes try to reallocate something that X was using as a new
pipe_resource, and potentially conflict in our rendering. But even
worse, if we reallocated the BO as a shader, the kernel would reject
rendering using the shader.
Not sure what happened in my testing that made the previous shadow
code fix glxgears swapbuffering, but this also fixes lots of CopyArea
in X (like dragging xlogo around in metacity).
I forgot to make the change in 96f164f6f0.
This fixes a warning with GCC and probably an error with Clang.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Without it, texcoords are mapped to GENERIC[0..7], PointCoord is mapped to
GENERIC[8], and user-defined varyings start from GENERIC[9]. Since texcoords
can only be used between VS and PS, and PointCoord is PS-only, it's silly to
always start from GENERIC[9] in all other shaders (such as LS, HS, ES, GS).
This adds support for TEXCOORD and PCOORD semantics. As a result, st/mesa
will use GENERIC[0] as a base for user-defined varyings, which should make
linking ES and GS as well as tessellation shaders at runtime easier.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
The problem is that the EDGEFLAG has to be toggled at vertex submission
time. This can be done from either the draw or the regular paths. Avoid
falling back to draw just because there's an edgeflag.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
Commit 8acaf862df switched things over to use TEXCOORD instead of
GENERIC, but did not update the nv30 swtnl draw paths. This teaches the
draw logic about TEXCOORD.
Among other things, this fixes a crash in demos/arbocclude when using
swtnl. Curiously enough, the point-sprite piglit works without this.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
These are only used once per draw, so it makes sense to keep them in
GART. Also take this opportunity to modernize the buffer mapping API
usage.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ben Skeggs <bskeggs@redhat.com>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
Instead of always having it in the data, let the bo placement decide it.
This fixes glxgears with swtnl forced on.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ben Skeggs <bskeggs@redhat.com>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
Apparently some compilers think we probably wanted to do !(x == y) instead
and issue a warning, so just shut it up... No functional change, obviously.
Cc: <mesa-stable@lists.freedesktop.org>
This fixes glxgears with NV30_SWTNL=1 forced on. Probably fixes a bunch
of other situations where we fall back to the swtnl path.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
nv30_validate_clip depends on the rasterizer state. Also we should
upload all the new clip planes on change since next time the plane data
won't have changed, but the enables might.
This fixes fixed-clip-enables and vs-clip-vertex-enables shader tests.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
Clearing can happen at a time when various state objects are incoherent
and not ready for a draw. Some of the validation functions don't handle
this well, so only flush the framebuffer state. This has the advantage
of also not doing extra work.
This works around some crashes that can happen when clearing.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
There can be scenarios where the "indirect" arg of a PFETCH becomes
known, and so the code will attempt to propagate it. Use this
opportunity to just fold it into the first argument, and prevent the
load propagation pass from touching PFETCH further.
This fixes gs-input-array-vec4-index-rd.shader_test and
vs-output-array-vec4-index-wr-before-gs.shader_test on nvc0 at least.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
The 1024 is already reported everywhere, not sure where this 0x1ff came
from.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
PIPE_QUERY_TIMESTAMP_DISJOINT could not work because q->ready was always
set to FALSE. To fix this issue, add more different states for queries
according to nvc0.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
We forgot to convert to VFETCH in case of indirect access. Fix that.
This avoids crashes on the new gs-input-array-vec4-index-rd and
vs-output-array-vec4-index-wr-before-gs but they still fail.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
When we get something like IN[ADDR[0].x+5], we will now guess that we
should look at IN[5] for the "base" information.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
In the case of a compare, the destination might be a predicate, but we
still want to flush denorms.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
This covers the pattern where a KILL_IF is used, which triggers a
comparison of -x to 0. This can usually be folded into the comparison whose
result is being compared to 0, however it may, itself, have already been
combined with another comparison. That shouldn't impact the logic of
this pass however. With this and the & 1.0 change, code like
00000020: 001c0001 80081df4 set b32 $r0 lt f32 $r0 0x3e800000
00000028: 001c0000 201fc000 and b32 $r0 $r0 0x3f800000
00000030: 7f9c001e dd885c00 set $p0 0x1 lt f32 neg $r0 0x0
00000038: 0000003c 19800000 $p0 discard
becomes
00000020: 001c001d b5881df4 set $p0 0x1 lt f32 $r0 0x3e800000
00000028: 0000003c 19800000 $p0 discard
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This has started to happen more now that the backend is producing
KILL_IF more often.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Tested on NVA8. No regression for ARB_pipeline_statistics piglit tests.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
A sequence number is written for 32-bits queries to make sure they are
ready, but not for 64-bits queries. Instead, we have to use a fence in
order to fix the HUD because it doesn't wait until the result is ready.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
v2: make this also compatible with original released firmware
v3 (chk): switch to original idea of separate files for fw versions
Signed-off-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v2)
Fixes non-determinism in bin/point-sprite rendering, and the stars on
the intro screen to neverball.
Cc: "10.6" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
get_immediate will return a const reference, the requested immediate
isn't necessarily in the x slot. Make sure to use the swizzle.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Previously, we used intrinsic->const_index[1] to represent "the number of
array elements to load" for load/store intrinsics. However, this set to 1
by every pass that ever creates a load/store intrinsic. Also, while it
might make some sense for registers, it makes no sense whatsoever in SSA.
On top of that, the i965 backend was the only backend to ever support it;
freedreno and vc4 just assert that it's always 1. Let's just delete it.
Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Rob Clark <robclark@freedesktop.org>
It's a remnant of some old NV extension. Unused.
I also have a patch that removes predicates if anyone is interested.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
This adds both ARB_texture_gather and the enhanced gather
for ARB_gpu_shader5.
This passes all the piglit tests, it relies on the GLSL
lowering pass to make textureGatherOffsets work.
v2: use inline to get gather component (Brian)
fix function name, add asserts (Brian)
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is a prep change for gather, and it makes more sense
to use an array in these cases.
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This was an oversight when GLSL1.30 was enabled, I think my
misunderstanding.
This fixes a bunch of tex-miplevel-selection tests under softpipe,
and is required for textureGather support.
I'm not sure this won't make sampling slowering, but its softpipe,
correctness first and all that.
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This moves some of the image filter args into a struct,
and passes that instead, this is prep work for adding texture
gather support which needs new arguments.
review: make filter args const.
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
A fence can outlive the ctx, so we shouldn't deref the ctx to get at the
screen. We need some updates in libdrm_freedreno API to completely
handle fences properly, but this is at least an improvement.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
This was causing corruption with hw binning on a306. Unlikely that it
is a306 specific, but rather the smaller gmem size resulted in different
tile configuration which was triggering the bug at certain resolutions.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Cc: "10.4" and "10.5" and "10.6" <mesa-stable@lists.freedesktop.org>
Whitelist adreno 306 (as found in msm8916/apq8016). Works pretty much
out of the box, although the smaller GMEM size requires more tiles to
fit 1920x1080, so bump up the max # of tiles as well.
Since it is just whitelist + trivial change, it makes sense to land on
all the active release branches.
Note that a305c ends up with gpu-id "306", hence a306 ends up with
gpu-id of "307". Apparently that is what happens when you let the
marketing dept name things.
Cc: "10.4" and "10.5" and "10.6" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Rob Clark <robclark@freedesktop.org>
These warnings have been detected by Clang 3.6.
codegen/nv50_ir_from_tgsi.cpp:1319:10: warning: struct 'Source' was
previously declared as a class [-Wmismatched-tags] const struct tgsi::Source *code;
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
The nv40_fp_bra() function in the same file is also unused but this is
the only place where the nv30/nv40 isa is documented.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This fixes a crash when trying to monitor MP counters because compute
support is not implemented for nvf0.
Reported-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Some bits were already there for texture views but some were missing.
In particular for cube map views things needed to change a bit.
For simplicity I ended up removing the separate face addr bit (just use
the z bit) - cube arrays didn't use it already, so just follow the same
logic there. (In theory using separate bits could allow for better hash
function but I don't think anyone ever did some measurements of that so
probably not worth the trouble, if we'd reintroduce it we'd certainly
wanted to use the same logic for cube arrays and cube maps.)
Also extend the seamless cube sampling to cube arrays - as there were no
piglit failures before this is apparently untested, but things now generally
work quite the same for cube textures and cube array textures so there
hopefully shouldn't be any trouble...
49 new piglits, 47 pass, 2 fail (both due to fake multisampling).
v2: incorporate Brian's feedback, add sampler view validation,
function rename, formatting fixes.
Reviewed-by: Brian Paul <brianp@vmware.com>
All the functionality was pretty much there, just not tested.
Trivially fix up the missing pieces (take target info from view not
resource), and add some missing bits for cubes.
Also add some minimal debug validation to detect uninitialized target values
in the view...
49 new piglits, 47 pass, 2 fail (both related to fake multisampling,
not texture_view itself). No other piglit changes.
v2: move sampler view validation to sampler view creation, update docs.
Reviewed-by: Brian Paul <brianp@vmware.com>
This aligns it to work similarly to nv50. However there's no library
code there, so the whole thing can be freed. Here we end up with an
allocated node that's not attached to a specific program.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86792
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
This avoids future addition to PIPE_PRIM_ from causing regressions
on r600g.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Our lower if/else pass was missed when converting NIR to use linked
lists rather than hashsets to track use/def sets.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Commit 44673512a8 enabled support for saturating fmul. However
experimentally this does not seem to work on the older chips. Restrict
the feature to G200 (NVA0) and later.
Reported-by: Pierre Moreau <pierre.morrow@free.fr>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90350
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Tested-by: Pierre Moreau <pierre.morrow@free.fr>
Reviewed-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Cc: mesa-stable@lists.freedesktop.org
Since we update num_vtxelts here, we could otherwise end up with stale
instancing information in the upper bits which wouldn't otherwise get
reset. (Also we run the risk of the previous draw having set the first
element as instanced.)
This appears as one of the causes for the test pointed out in fdo#90363
to fail on nvc0.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90363
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
See identical commit for nv50. Destroying the current context and then
creating a new one or switching to another existing context would cause
the "current" state to not be properly initialized, so we save it off in
the screen.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Normally this is kept in nv50_context, and on switching the active
context, the state is copied from the previous context. However when the
last context is destroyed, this is lost, and a new context might later
be created. When the currently-active context is destroyed, save its
state in the screen, and restore it when setting the current context.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90363
Reported-by: Matteo Bruni <matteo.mystral@gmail.com>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Tested-by: Matteo Bruni <matteo.mystral@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
The linked list in gallium is pretty much the kernel list and we would like
to have a C-based linked list for all of mesa. Let's not duplicate and
just steal the gallium one.
Acked-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Rob Clark <robclark@freedesktop.org>
The former logic would copy the saturate up to any mul with an immediate
if there was a subsequent mul with a saturate. However we only want to
do that if we collapsed 2 muls by multiplying their immediates (or were
able to put the immediate in as a post-multiplier).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Having the wrong inferred type prevents a number of optimizations,
including constant propagation (since float immediates work differently
than integer immediates).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Fix Clang return-type error introduced with commit
96f164f6f0 "gallium: make
pipe_context::begin_query return a boolean".
CC r600_query.lo
r600_query.c:443:3: error: non-void function 'r600_begin_query' should return a value [-Wreturn-type]
return;
^
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Martin Peres <martin.peres@free.fr>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Martin Peres <martin.peres@free.fr>
This patch defines "Driver statistics" and "MP counters" groups, but
only the latter will be exposed through GL_AMD_performance_monitor.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Martin Peres <martin.peres@free.fr>
GL_AMD_performance_monitor must return an error when a monitoring
session cannot be started.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Martin Peres <martin.peres@free.fr>
This allows queries to return different numeric types.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Martin Peres <martin.peres@free.fr>
According to the spec of GL_AMD_performance_monitor, valid type values
returned are UNSIGNED_INT, UNSIGNED_INT64_AMD, PERCENTAGE_AMD, FLOAT.
This also introduces the new field group_id in order to categorize
queries into groups.
v2: add PIPE_DRIVER_QUERY_TYPE_BYTES
v3: fix incorrect query type for radeon and svga drivers
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Martin Peres <martin.peres@free.fr>
src1 would contain the predicate, which would get emitted as a register
source by an undiscerning srcId helper. Work around this in the same way
as in emitTEX.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
This was causing src0 to always have the absolute value flag set.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Accidentally added since the introduction of the file.
Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This will set the FTZ flag (flush denorms to zero) on all opcodes that
can take it.
This resolves issues in Unigine Heaven 4.0 where there were solid-filled
boxes popping up.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89455
Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
When there is a colormask active that does not cover all the channels,
enable reading in the destination like with a combining blend
operation. This fixes fbo-blending-formats on a3xx.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Enables ARB_depth_buffer_float. There is no sampling support for
interleaved Z32F_S8, so we store the two textures separately, one as
Z32F, the other as S8. As a result, we need a lot of additional logic
for restores and transfers.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
32-bit depth buffers are stored as unorm, and thus need special handling
when moving to and from gmem. They are copied into gmem by writing
depth, and resolved from gmem using a special resolve bit which
apparently float-ifies the data.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
This fixes a number of regressions since
61393bdcdc
u_tile: fix stencil texturing tests under softpipe
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89960
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The bit to enable .z is still commented out, as it is triggering gpu
hangs in 0ad. But at least gl_FragCoord.w works now, and we know what
bits we are *supposed* to set for .z (with that uncommented all piglit
fragcoord tests are passing).
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Update formats table with new formats that Ilia has figured out, and fix
sampling from srgb texture and integer vbo's.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
... to manage the LIBDRM*_CFLAGS. The former is the recommended approach
by the Android build system developers while the latter has been
depreciated for quite some time.
Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
isaml needs to scale up coords based on LoD. Also fix bogus bary.f
varying # when there are non-bary frag shader inputs. And use sub.s of
a positive immediate rather than add.s of negative (since CP is better
about figuring out that those can be collapsed into the cat2 instr).
Signed-off-by: Rob Clark <robclark@freedesktop.org>
For now, completely flatten if/else blocks. That will almost certainly
change once we have flow control.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Basically just sync up the cmdstream emit parts to match the changes
already done on a3xx.
Also, fix scheduling for mem instructions. This is needed on a4xx, and
I am a bit surprised it isn't needed for a3xx.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
As mentioned by Michel Dänzer for LLVM >= 3.6 we create the
LLVMTargetMachine (with triple amdgcn--), as we setup the radeonsi
context. For older LLVM or hardware (r600) the triple is always r600--
and is created at a later stage - radeon_llvm_compile()
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
We could potentially support the right combination of 8888 to 565, but the
important thing for now is to not mix up our orderings of 8888. Fixes
fbo-copyteximage regressions.
For blitting, we want to fire off an RCL-only job. This takes a bit of
tweaking in our validation and the simulator support (and corresponding
new code in the kernel).
I want to be able to have multiple jobs being set up at the same time (for
example, a render job to do a little fixup blit in the course of doing a
render to the main FBO).
So, it turns out my simulator doesn't *quite* match the hardware. And the
errata about raster textures tells you most of what's wrong, but there's
still stuff wrong after that. Instead, if we're asked to sample from
raster, we'll just blit it to a tiled temporary.
Raster textures should only be screen scanout, and word is that it's
faster to copy to tiled using the tiling engine first than to texture from
an entire raster texture, anyway.
These are required to get piglit's idiv tests working. The
unsigned<->float conversions are wrong, but are good enough to get
piglit's small ranges of values working.
We're over-allocating our BCL in vc4_draw.c, so this never mattered.
However, new RCL-only blit support might end up here without having set up
any BCL contents.
Just build up arrays for src0/src1, and use create_collect()..
Also add back missing .3d flag for 3d/cube textures.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
I noticed some cases where we where trying to copy-propagate indirect
src's into places they cannot go, like 2nd src for cat3 (mad, etc).
Expand out valid_flags() to be aware of relativ flag, and fix up a few
related spots.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
When we get in a scenario where we cannot schedule any more instructions
due to address register conflict, clone the instruction that writes the
address register, and switch the remaining unscheduled users for the
current address register over to the new clone.
This is simpler and more robust than the previous attempt (which tried
and sometimes failed to ensure all other dependencies of users of the
address register were scheduled first).. hint it would try to schedule
instructions that were not actually needed for any output value.
We probably need to do the same with predicate register, although so far
it isn't so heavily used so we aren't running into problems with it
(yet).
Signed-off-by: Rob Clark <robclark@freedesktop.org>
A bit fugly.. try and make this cleaner.. note if we hoist all the
get_addr() out of the loop we can drop the hashtable and just use
create_addr()..
Signed-off-by: Rob Clark <robclark@freedesktop.org>
It probably *should* be an assert, but for now TGSI f/e isn't very good
about dealing w/ CONST vs ABS/NEG. So for debug builds, print a warning
instead of crashing with an assert for now.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Without this, a3xx breaks.. a4xx would too if it had already implemented
support for passing driver params.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
For a normal MAD (ie. not MADSH), if first source is gpr and second
source is const, we can swap the first two sources to avoid needing a
mov instruction.
This gives back the biggest advantage TGSI f/e had over NIR f/e for
common shaders, since TGSI f/e had this logic in the f/e. Note that
doing this in copy-prop step has the advantage that it will also work
for cases like:
MOV TEMP[b], CONST[x]
MAD TEMP[d], TEMP[a], TEMP[b], TEMP[c]
Signed-off-by: Rob Clark <robclark@freedesktop.org>
llvm goes crazy when doing that, using way more memory and time, though there's
probably more to it - this points to a very much similar issue as fixed in
8a9f5ecdb1. In any case I've seen a quite
plain looking vertex shader with just ~50 simple tgsi instructions (but with a
dozen or so such indirect constant buffer lookups) go from a terribly high
~440ms compile time (consuming 25MB of memory in the process) down to a still
awful ~230ms and 13MB with this fix (with llvm 3.3), so there's still obvious
improvements possible (but I have no clue why it's so slow...).
The resulting shader is most likely also faster (certainly seemed so though
I don't have any hard numbers as it may have been influenced by compile times)
since generally fetching constants outside the buffer range is most likely an
app error (that is we expect all indices to be valid).
It is possible this fixes some mysterious vertex shader slowdowns we've seen
ever since we are conforming to newer apis at least partially (the main draw
loop also has similar looking conditionals which we probably could do without -
if not for the fetch at least for the additional elts condition.)
v2: use static vars for the fake bufs, minor code cleanups
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Add SV_GEOMETRY_EMIT special variable type to track the
implicit dependencies between CUT/EMIT_VERTEX/MEM_RING
instructions so GCM/scheduler doesn't reorder them.
Mark emit instructions as unkillable so DCE doesn't eat them.
Enable only for evergreen/cayman as there are a few
unexplained GS piglit regressions on R6xx/R7xx with SB
enabled otherwise.
Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
CF_END could end up emitted in the middle of a shader on cayman
when there was a loop at the very end.
Fixes glsl-1.50-geometry-end-primitive and
ext_transform_feedback-geometry-shaders-basic piglit tests.
Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We limit y-tiling to 0x20 when depth is involved. However the function is
run for each miplevel, and the hardware expects miplevel 0 to have the
highest tiling settings. Perform the y-tiling limit on all levels of a
3d texture, not just the ones that have depth.
Fixes:
texelFetch fs sampler3D 98x129x1-98x129x9
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Tested-by: Nick Tenney <nick.tenney@gmail.com> # GT216
Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
This code to handle absolute values on op3 srcs was a bit too simple,
it really needs a temp reg per src, not one per channel, make it
easier and let sb clean up the mess.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89831
Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The NIR compiler frontend is an alternative to the TGSI f/e, producing
the same ir3 IR and using the same backend passes for scheduling, etc.
It is not enabled by default yet, as there are still some regressions.
To enable, use 'FD_MESA_DEBUG=nir'. It is enough to use with, for
example, xonotic or supertuxkart.
With the NIR f/e, scalarizing and a number of other lowering steps
happen in NIR, so we don't have to do them in ir3. Which simplifies the
f/e and allows the lowered instructions to pass through other
optimization stages.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Use the correct sprite replacement depending on the flip of the coord
mode, using either T or 1-T depending on whether we have an upper-left or
lower-left coordinate origin. This fixes all the point sprite piglits.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Copies nouveau_buffer and radeon_buffer. This allows a write to proceed
to an uninitialized part of a buffer even when the GPU is using the
previously-initialized portions.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Waiting on a bo being ready is handled in fd_bo_cpu_prep. No need to
keep separate timestamps around.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
A resource flush is an upload of a hypothetically-staging texture to the
GPU. For a UMA system, this will largely be a no-op or
cache-maintenance. Move the render flush logic into transfer_map where
it belongs, and clear out the transfer_flush function.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
pipe_sampler_view already contains a texture, remove the redundant
tex_resource member which pointed at the same thing.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Since NIR f/e currently encodes immediates in instructions (rather than
passing via const), we need to ensure that when const's are used the get
initialized to the proper values. Otherwise comparing NIR to TGSI
compiler, it will use proper immediate values in one case, and randomly
initialize values in the other. Which confuses ir3test.
Signed-off-by: Rob Clark <robclark@freedesktop.org>