This fixes a bunch of unnecessary barriers due to the scheduler not
knowing what that arbitrary register description refers to when trying
to reason about its dependencies.
The result is rescheduling in the convolution kernel shader in
Lightsmark, which results in avoiding register spilling and increasing
the performance of the first scene from 6-7 fps midway through the
panning to 11fps. The register spilling was a regression from Mesa
7.9 to Mesa 7.10.
Improves performance of my GLSL demo by 5.1% (+/- 1.4%, n=7). It also
reschedules the giant multiply tree at the end of
glsl-fs-convolution-1 so that we end up not spilling registers,
producing the expected level of performance.
Can happen during swrast fallbacks if a buffer is somehow bound as
a render target and a texture.
Fixes gnome-shell on nv20, and gets it mostly working on nv10.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
sw clears were being used and not getting the correct offsets in the span
code.
also not emitting correct offsets for CB draws to texture levels.
(I've no idea why I'm playing with r100).
This is a candidate for 7.9 and 7.10
According to R700 ISA we have only two channels for cfile constants.
This patch makes piglit tests "glsl1-constant array with constant
indexing" happy on RV710.
Fixes the following Piglit tests:
glslparsertest/shaders/array2.frag
glslparsertest/shaders/dataType6.frag
NOTE: This is a candidate for the 7.9 and 7.10 branches.
This fixes a potential failure when a begin/end_query is the first
thing to happen after flushing the scene.
NOTE: This is a candidate for the 7.10 and 7.9 branches.
This revealed a bug in ra_get_spill_benefit where we only considered
the benefit of the first adjacency we were to remove, explaining some
of the ugly spilling I've seen in shaders. Because of the reduced
spilling, it reduces the runtime of glsl-fs-convolution-1 36.9% +/-
0.9% (n=5).
This was recommended in the original paper, but I figued "make it run"
before "make it fast". Now we make it fast. Reduces the runtime of
glsl-fs-convolution-1 by 12.7% +/- 0.6% (n=5).
Our use of the register allocator in i965 is somewhat unusual.
Whereas most architectures would have a smaller set of registers with
fewer register classes and reuse that across compilation, we have 1,
2, and 4-register classes (usually) and a variable number up to 128
registers per compile depending on how many setup parameters and push
constants are present. As a result, when compiling large numbers of
programs (as with glean texCombine going through ff_fragment_shader),
we spent much of our CPU time in computing the q[] array. By keeping
a separate list of what the conflicts are for a particular reg, we
reduce glean texCombine time 17.0% +/- 2.3% (n=5).
We don't expect this optimization to be useful for 915, which will
have a constant register set, but it would be useful if we were switch
to this register allocator for Mesa IR.
Fixes texrect-many regression with ff_fragment_shader -- as we added
refs to the subsequent texcoord scaling paramters, the array got
realloced to a new address while our params[] still pointed at the old
location.
The driver was saying that independend blend functions was not supported,
but it really was. The driver was using the per-target independend blend
factors but the state tracker was only setting the 0th one (per the
Gallium spec).
Fixes a piglit fbo-drawbuffers2-blend regression.
See https://bugs.freedesktop.org/show_bug.cgi?id=33215
The removed semantic check also exists in ast_type_specifier::hir(), which
is a more natural location for it.
The check verified that precision statements are applied only to types
float and int.
* Check that precision qualifiers only appear in language versions 1.00,
1.30, and later.
* Check that precision qualifiers do not apply to bools and structs.
Fixes the following Piglit tests:
* spec/glsl-1.30/precision-qualifiers/precision-bool-01.frag
* spec/glsl-1.30/precision-qualifiers/precision-struct-01.frag
* spec/glsl-1.30/precision-qualifiers/precision-struct-02.frag
