| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
| |
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Reduces compiled size of brw_wm_surface_state.o another 1.9%.
Overall, this brw_wm_surface_state reduction series cuts
firefox-talos-gfx runtime by 0.68% +/- 0.42% (n=6).
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
| |
This reduces compiled size (4.7% of brw_wm_surface_state.o).
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
| |
This massively reduces compiled size (6.7% of brw_wm_surface_state.o).
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
| |
This massively reduces compiled size (4.9% of brw_wm_surface_state.o).
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
| |
It turns out that gcc is just awful at generating code for
brw_structs.h style state setup, and using bitshifting on u32s
generates better code while being similarly readable (and more
verifiable compared to the specs, using the INTEL_MASK macro).
|
|
|
|
|
|
|
|
| |
It's only used in the old fragment program path, to avoid projection
when w is always 1. We do want to do this in the new path pre-gen6
too, but we'll probably do it through the ir.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
| |
Oddly, this increases compiled code size. (marking the 'if' as likely
also increases code size, but not as much).
Reviewed-by: Kenneth Graunke <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
| |
Interestingly, the compiler wasn't doing this for us at -O2, so we
were doing the computation for every non-_ReallyEnabled unit.
Reviewed-by: Kenneth Graunke <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
| |
Reviewed-by: Eric Anholt <[email protected]>
|
|
|
|
| |
The pitch of 4 is allowed for squared microtiling only.
|
|
|
|
|
|
|
|
|
| |
- all asics need to emit CONTEXT_CONTROL
- all r6xx asics need to emit 3D_START_CMDBUF
The ddx and r600c already do this. r600g should as well.
Signed-off-by: Alex Deucher <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
We are getting inconsistent methods for endian detection (same answer when
it works, just doesn't work on some platforms) depending on whether __GLIBC__
is defined, which of course depends on include ordering before p_config.h
Just make p_config.h include limits.h to solve this.
Signed-off-by: Dave Airlie <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
| |
On my original R600 card this at least lets gnome shell run for a while longer
and the piglit r300-readcache test case works a lot more reliably.
Still a few more stability issues running a piglit test run though.
Signed-off-by: Dave Airlie <[email protected]>
|
|
|
|
|
|
| |
The original R600 doesn't have these so don't emit them.
Signed-off-by: Dave Airlie <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The spec doesn't state it should be an error, but. We have this piglit test
useprogram-inside-begin that passes with this commit. No idea what's correct.
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
The conditional rendering should be able to kill CopyPixels.
I assume the render condition has no effect on resource_copy_region.
This fixes piglit:
- NV_conditional_render/copypixels
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
| |
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
| |
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
| |
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
| |
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
| |
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Always default to DEFAULT_*_FORMATS for mandatory GL formats.
(st_choose_format must not fail for those)
Use DEFAULT_RGBA when alpha is required instead of RGB.
Use DEFAULT_RGB otherwise.
These are more or less the remaining differences between the old code and
the new one.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
| |
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
| |
This makes D32 work again on chipsets which can't do it.
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The problem is: The second time the function is called with a new
internal format, strb->format is usually not PIPE_FORMAT_NONE.
RenderbufferStorage(... GL_RGBA8 ...);
RenderbufferStorage(... GL_RGBA16 ...); // had no effect on the format
Broken with: fd6f2d6e5783d8810d0ab88e1c470958fd5eb2eb
Test: piglit/fbo-storage-completeness
NOTE: This is a candidate for the 7.10 branch.
(if fd6f2d6e5783d8810d0ab88e1c470958fd5eb2eb is cherry-picked as well)
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
Lowered indirect addressing can create lots of immediates.
Fixes piglit/glsl-fs-uniform-array-7 on r300g.
NOTE: This is a candidate for the 7.10 branch.
Reviewed-by: Brian Paul <[email protected]>
|
| |
|
| |
|
|
|
|
|
| |
Fixes piglit:
- NV_conditional_render/clear
|
|
|
|
|
| |
This is just a temporary solution for now until there is a better way
to share code between mesa and gallium.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
From now on, depth test is always enabled in hardware.
If depth test is disabled in Gallium, the hardware Z function is set to ALWAYS.
If there is no zbuffer set, the colorbuffer0 memory is set as a zbuffer
to silence the CS checker.
This fixes piglit:
- occlusion-query-discard
- NV_conditional_render/bitmap
- NV_conditional_render/drawpixels
- NV_conditional_render/vertex_array
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
We want to check for Success, otherwise it will fail even with the right visual.
NOTE: This is a candidate for the 7.10 branch.
Signed-off-by: Antoine Labour <[email protected]>
Signed-off-by: Stéphane Marchesin <[email protected]>
Signed-off-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
When using _mesa_layout_parameters, all params copied in the 'layout'
output in the PASS 1 don't modify StateFlags (because they are simply
memcpy'ed).
This patch fixes the problem, assuring output gl_prog_param_list
StateFlags field is the same as the input one.
NOTE: This is a candidate for the 7.10 branch.
Signed-off-by: Brian Paul <[email protected]>
|
|
|
|
| |
I failed to commit and squash before pushing.
|
|
|
|
|
|
| |
This makes automatic parsing of MESA_GLSL=dump output easier.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
| |
At glLinkShaders time, a fail() call in FS compile in 8-wide (the one
that's required to succeed, though we may relax that at some point for
pre-Ironlake performance) will now report out as a link error.
|
|
|
|
|
|
|
|
|
| |
We now have:
brw_fs.cpp handles calling out to everything and optimization.
brw_fs_visitor.cpp handles translating to our LIR.
brw_fs_emit.cpp handles emitting from our LIR to native code.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
| |
This is all separate from the visitor and the optimization passes
which feed into it.
Reviewed-by: Kenneth Graunke <[email protected]>
|