Brian Paul
2996ce72b1
llvmpipe: add a cast
2010-11-02 11:53:14 -06:00
Brian Paul
9fbf744389
llvmpipe: assign context's frag shader pointer before using it
...
The call to draw_bind_fragment_shader() was using the old fragment
shader. This bug would have really only effected the draw module's
use of the fragment shader in the wide point stage.
2010-11-02 11:50:37 -06:00
José Fonseca
85a08f8fc7
gallivm: Remove the EMMS opcodes.
...
Unnecessary now that lp_set_target_options() successful disables MMX code
emission.
2010-10-28 20:42:02 +01:00
Keith Whitwell
0072acd447
Merge remote branch 'origin/master' into lp-setup-llvm
...
Conflicts:
src/gallium/drivers/llvmpipe/lp_setup_coef.c
src/gallium/drivers/llvmpipe/lp_setup_coef.h
src/gallium/drivers/llvmpipe/lp_setup_coef_intrin.c
src/gallium/drivers/llvmpipe/lp_setup_point.c
src/gallium/drivers/llvmpipe/lp_setup_tri.c
src/gallium/drivers/llvmpipe/lp_state_derived.c
src/gallium/drivers/llvmpipe/lp_state_fs.h
2010-10-17 19:09:42 -07:00
José Fonseca
a0add0446c
llvmpipe: Fix bad refactoring.
...
'i' and 'chan' have random values here, which could cause a buffer
overflow in debug builds, if chan > 4.
2010-10-17 09:58:04 -07:00
Keith Whitwell
ac98519c4e
llvmpipe: validate color outputs against key->nr_cbufs
2010-10-15 14:49:13 +01:00
Keith Whitwell
ffab84c9a2
llvmpipe: check shader outputs are non-null before using
2010-10-15 14:49:13 +01:00
Keith Whitwell
0a1c900103
llvmpipe: don't pass frontfacing as a float
2010-10-15 13:27:47 +01:00
Brian Paul
3d7479d705
llvmpipe: code to dump bytecode to file (disabled)
2010-10-14 17:28:24 -06:00
Keith Whitwell
f0bd76f28d
llvmpipe: don't try to emit non-existent color outputs
2010-10-14 14:08:20 +01:00
José Fonseca
95c18abb03
llvmpipe: Unbreak Z32_FLOAT.
...
Z32_FLOAT uses <4 x float> as intermediate/destination type,
instead of <4 x i32>.
The necessary bitcasts got removed with commit
5b7eb868fd
Also use depth/stencil type and build contexts consistently, and
make the depth pointer argument a ordinary <i8 *>, to catch this
sort of issues in the future (and also to pave way for Z16 and
Z32_FLOAT_S8_X24 support).
2010-10-13 15:25:15 +01:00
José Fonseca
986cb9d5cf
llvmpipe: Use lp_tgsi_info.
2010-10-11 13:06:25 +01:00
José Fonseca
307df6a858
gallivm: Cleanup the rest of the flow module.
2010-10-09 21:39:14 +01:00
José Fonseca
d45c379027
gallivm: Remove support for Phi generation.
...
Simply rely on mem2reg pass. It's easier and more reliable.
2010-10-09 20:14:03 +01:00
José Fonseca
cc40abad51
gallivm: Don't generate Phis for execution mask.
2010-10-09 12:55:31 +01:00
Keith Whitwell
5b7eb868fd
llvmpipe: clean up shader pre/postamble, try to catch more early-z
...
Specifically, can do early-depth-test even when alpahtest or
kill-pixel are active, providing we defer the actual z write until the
final mask is avaialable.
Improves demos/fire.c especially in the case where you get close to
the trees.
2010-10-09 11:44:45 +01:00
Keith Whitwell
aa4cb5e2d8
llvmpipe: try to be sensible about whether to branch after mask updates
...
Don't branch more than once in quick succession. Don't branch at the
end of the shader.
2010-10-09 11:44:45 +01:00
Keith Whitwell
d2cf757f44
gallivm: specialized x8z24 depthtest path
...
Avoid unnecessary masking of non-existant stencil component.
2010-10-09 11:44:09 +01:00
Keith Whitwell
954965366f
llvmpipe: dump fragment shader ir and asm when LP_DEBUG=fs
...
Better than GALLIVM_DEBUG if you're only interested in fragment shaders.
2010-10-09 11:43:23 +01:00
Keith Whitwell
40d7be5261
llvmpipe: use alloca for fs color outputs
...
Don't try to emit our own phi's, let llvm mem2reg do it for us.
2010-10-09 11:43:23 +01:00
Keith Whitwell
8009886b00
llvmpipe: defer attribute interpolation until after mask and ztest
...
Don't calculate 1/w for quads which aren't visible...
2010-10-09 11:42:48 +01:00
José Fonseca
1c32583581
gallivm: Only apply min/max_lod when necessary.
2010-10-06 18:50:57 +01:00
José Fonseca
446dbb9217
llvmpipe: Dump a few missing shader key flags.
2010-10-06 11:41:08 +01:00
Keith Whitwell
591e1bc34f
llvmpipe: make debug_fs_variant respect variant->nr_samplers
2010-10-06 11:40:30 +01:00
José Fonseca
87267c71f6
llvmpipe: Make rgb/alpha bland func/factors match, when there is no alpha.
...
Makes AoS blending easier, and state more canonical.
2010-09-22 15:02:39 +01:00
José Fonseca
b556bb7c44
llvmpipe: When failing free fs shader too.
2010-09-21 17:51:29 +01:00
Brian Paul
955d76c3d2
llvmpipe: maintain fragment shader state for draw module
2010-09-20 12:52:16 -06:00
Keith Whitwell
5b4c43d985
llvmpipe: use llvm for attribute interpolant calculation
...
Basically no change relative to hard-coded version, but this will
be useful for other changes later.
2010-09-18 08:40:17 +01:00
Keith Whitwell
0986355425
llvmpipe: add DEBUG_FS to dump variant information
2010-09-16 17:34:58 +01:00
Vinson Lee
aeb83928fd
llvmpipe: Initialize variable for potentially unhandled switch case.
2010-09-14 00:17:13 -07:00
Keith Whitwell
67b957781d
llvmpipe: pass linear masks to fragment shader
...
Fragment shader can extract the correct bits for each quad.
2010-09-12 15:03:49 +01:00
José Fonseca
853953dc3c
llvmpipe: Only generate the whole shader specialization for opaque shaders.
...
If not opaque, then the color buffer will have to be read any way,
therefore the specialization is pointless.
2010-09-12 10:15:48 +01:00
José Fonseca
277d768d10
llvmpipe: Don't store display the alpha ref value in the key.
...
It's never used.
2010-09-11 13:30:14 +01:00
Brian Paul
3ec296390c
llvmpipe: remove redundant tgsi_dup_tokens() call
...
We were calling this twice so the first allocation was orphaned/leaked.
2010-09-09 10:04:39 -06:00
Keith Whitwell
9f6e8e1d6b
llvmpipe: use opcodes instead of function pointers in bins
...
Also, move some state from rasterizer struct to the scene.
2010-09-07 14:02:15 +01:00
Keith Whitwell
71e27ef21c
llvmpipe: put fs variant dumping in a function
2010-09-07 13:22:54 +01:00
José Fonseca
00989d5bfc
llvmpipe: Relax the colormask constraint on opaque.
...
Also, include the color buffer in the key. Not having it there
causes a tight knots in the logic to determine when it is OK or not
to discard previous color buffer contents.
2010-09-05 10:17:51 +01:00
Keith Whitwell
b6e03eafe3
llvmpipe: fence debugging, add llvmpipe_finish
2010-08-25 10:29:27 +01:00
Keith Whitwell
3d4b60f1f7
llvmpipe: reduce size of fragment shader variant key
...
Don't spend as much time comparing them.
2010-08-22 14:49:17 +01:00
Vinson Lee
078eff659a
llvmpipe: Fix implicit declaration of lp_func_delete_body warnings.
2010-07-30 11:48:48 -07:00
Zack Rusin
042018a943
llvmpipe: delete function bodies after generating machine code
2010-07-30 16:25:29 +01:00
José Fonseca
e3d2ebac11
llvmpipe: Avoid corrupting the FPU stack with MMX instructions on 32bit OSes.
...
Unfortunately LLVM doesn't emit EMMS itself, and there is no
easy/effective way to disable MMX.
http://llvm.org/bugs/show_bug.cgi?id=3287
2010-07-29 19:10:59 +01:00
Keith Whitwell
d4b64167b5
llvmpipe: pass mask into fragment shader
...
Move this code back out to C for now, will generate separately.
Shader now takes a mask parameter instead of C0/C1/C2/etc.
Shader does not currently use that parameter and rasterizes whole
pixel stamps always.
2010-07-13 17:23:49 +01:00
Zack Rusin
b5e381d978
llvmpipe: make geometry shaders and stream output work
2010-06-22 12:58:04 -04:00
José Fonseca
197636814e
llvmpipe: s/lpfs/shader/
2010-06-21 14:11:15 +01:00
José Fonseca
d7f4250333
llvmpipe: Fix pointer dereference after free.
2010-06-18 17:37:37 +01:00
Roland Scheidegger
17c9d7eea7
llvmpipe: limit the number of fragment shader variants kept around
...
llvmpipe can create a large number of shader variants for a single shader
(which are quite big), and they were only ever deleted if the shader itself
was deleted. This is especially apparent in things like glean
blendFunc where a new variant is created for every different subtest, chewing
up all memory.
This change limits the numbers of fragment shader variants (for all shaders)
which are kept around to a fixed number. If that would be exceeded a fixed
portion of the cached variants is deleted (since without tracking the used
variants this involves flushing we don't want to delete only one).
Always the least recently used variants (from all shaders together) are
deleted.
For now this is all per-context.
Both the number of how many variants are cached (1024) as well as how many
will be deleted at once (1/4 of the cache size) are just rough guesses and
subject to further optimization.
2010-06-18 13:52:17 +01:00
Keith Whitwell
a6d9d18fae
llvmpipe: hook up basic gs and multiple constant buffer support
2010-06-07 16:39:02 +01:00
José Fonseca
81862bbbc8
llvmpipe: Number shaders/variants for more reproducible function names in the IR.
2010-06-03 14:32:56 +01:00
José Fonseca
147dc2354c
llvmpipe: Centralize all position interpolation in lp_bld_interp.c.
2010-06-02 13:24:25 +01:00