Mesa constant-folds built-in functions by using a miniature GLSL
interpreter (see
ir_function_signature::constant_expression_evaluate_expression_list()).
This interpreter had a bug in its handling of array indexing, which
caused expressions like "m[i][j]" (where m is a matrix) to be handled
incorrectly. Specifically, it incorrectly treated j as indexing into
the whole matrix (rather than indexing just into the vector m[i]); as
a result the offset computed for m[i] was lost and m[i][j] was treated
as m[j][0].
Fixes piglit tests inverse-mat[234].{vert,frag}.
NOTE: This is a candidate for the 9.1 and 9.0 branches.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=57436
Conceptually the same as previously done in float_to_half.
Should cut down number of instructions from 14 to 10 or so, but
will promote some NaNs to Infs, so it's disabled.
It gets a bit tricky though handling all the cases correctly...
Passes basic tests either way (though there are no tests testing special
cases, but some manual tests injecting them seemed promising).
v2: style and comment fixes suggested by Jose
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This replaces the existing float-to-half implementation.
There are definitely a couple of differences - the old implementation
had unspecified(?) rounding behavior, and could at least in theory
construct Inf values out of NaNs. NaNs and Infs should now always be
properly propagated, and rounding behavior is now towards zero
(note this means too large but non-Infinity values get propagated to max
representable value, not Infinity).
The implementation will definitely not match util code, however (which
does nearest rounding, which also means too large values will get
propagated to Infinity).
Also fix a bogus round mask probably leading to rounding bugs...
v2: fix a logic bug in handling infs/nans.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reduced stack size allows to run more threads in some cases,
improving performance for the shaders that use stack (that is, for the
shaders with control flow instructions). E.g. with unigine-based apps.
v4: implement exact computation taking into account wavefront size
v5: add cases for RV620, RS880
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
v2: reduce key size, don't copy key around to much.
v3: remove key size reduction
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This works different than on R600, we need to add the start instance manually.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
This should be used by both SI and R600.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Fixes mplayer -vo vdpau OSD.
NOTE: This is a candidate for the 9.1 branch.
Reported-by: Igor Vagulin <igor.vagulin@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Tested-by: Christian König <christian.koenig@amd.com>
A fix for lower_jumps progress reporting, very much like similar in
c1e591eed.
NOTE: This is a candidate for stable branches.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
All the other expression types allowed here have inst->mlen == 0, and this
one has implied MRF writes for all of its payload, so nothing else in the
implementation should need to change.
Reduces SEND messages for loading from pull constants in kwin's Lanczos
shader from 16 to 6. (Due to a deficiency in constant propagation, I
can't use the hack I did in the previous commit to test the performance
change)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=61554
NOTE: This is a candidate for the 9.1 branch.
This comes at a minor performance cost at the moment (-3.2% +/- 0.2%, n=14 on
my GM45 forced to load all uniforms through the varying-index path), but we
get a whole vec4 at a time to reuse in the next commit.
v2: Fix comment about channels in the other message.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
NOTE: This is a candidate for the 9.1 branch.
We weren't setting needs_dep[i] in the loops, so we'd continue on to
potentially add the same workaround MOVs to the later basic block
boundaries, too. We can either set needs_dep[i] to exit through the
normal path, or we can just return since we know we're done.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
For sampler messages, it depends on the target gen, and on gen4
SIMD16-sampler-on-SIMD8-execution we were returning 4 instead of 8 like we
should.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
NOTE: This is a candidate for the 9.1 branch.
I think this makes it much more obvious what's going on here.
NOTE: This is a candidate for the 9.1 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This is our first CSE on a regs_written() > 1 instruction, so it takes a
bit of extra fixup. Reduces the number of loads on kwin's Lanczos shader
from 12 to 2.
v2: Fix compiler warning (false positive on possibly-uninitialized variable)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=61554
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)
NOTE: This is a candidate for the 9.1 branch.
Like we have done for the VS and for constant-index uniform loads, we use
the sampler engine to get caching in front of the L3 to avoid tickling the
IVB L3 bug. This is also a bit of a functional change, as we're now
loading a vec4 instead of a single dword, though we're not taking
advantage of the other 3 components of the vec4 (yet).
With the driver hacked to always take the varying-index path for all
uniforms, improves performance of my old GLSL demo by 315% +/- 2% (n=4).
This a major fix for some blur shaders in compositors from the
varying-index uniforms support I introduced in 9.1.
v2: Move old offset computation into the pre-gen7 path.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=61554
NOTE: This is a candidate for the 9.1 branch.
Right now we don't have anything with regs_written() > 1 and !inst->mlen,
but that's about to change.
NOTE: This is a candidate for the 9.1 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We want to load vec4s, since loading a vec4 instead of a dword is
basically no increased latency. But for variable indexed access, the
previous requirement of aligned vec4s for a sampler LD was hard to
implement.
Note that this change only affects those messages that use the surface
format, like sampler LDs, but not to the untyped data cache loads we've
used in other cases.
No significant performance difference on my GLSL demo with uniforms forced
to take the varying pull constants path (n=4).
NOTE: This is a candidate for the 9.1 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This puts the rounding-up logic into the function itself instead of all
the callers having to manage it. Also drop an "unused" comment in gen4,
as the stride *is* used for texbos (and will be for uniforms soon).
NOTE: This is a candidate for the 9.1 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
I'm going to want to change the math for gen7 using sampler LD
instructions in a way that gets CSE to occur like we'd hope.
NOTE: This is a candidate for the 9.1 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We weren't inserting it into the list, so it did nothing. This line was
replaced by the MOV/MUL block above.
NOTE: This is a candidate for the 9.1 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This happens quite a bit with varying-index uniform loads. We could also
do better by avoiding the MACH entirely, but there's no reason not to at
least take this step.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If the active uniform is an array, then the length of the uniform name should
include the three extra characters for the "[0]" suffix, which is required by
the GL 4.2 spec to be appended to the uniform name in glGetActiveUniform().
This avoids the situation where the output buffer does not have enough space
to hold the "[0]" suffix, resulting in an incomplete array specification like
"foobar[0".
NOTE: This is a candidate for the 9.1 branch.
Change-Id: I41e87ba347a7169eec8c575596cc3416adbe0728
Signed-off-by: Haixia Shi <hshi@chromium.org>
Reviewed-by: Stéphane Marchesin <marcheu@chromium.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This is a more aggressive version of the old brw_optimize() path. Reduces
cycles spent in the vertex shader on minecraft by 18.6% +/- 10.0% (n=15).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We dump shader source in ir_to_mesa.cpp, and we dump linked programs here,
but we had no reference from the linked programs to their source. This
was preventing improvement of shader-db to use linked shader programs
instead of individual shader files (which is bogus, because it means we
optimize out VS outputs, and don't interpolate FS inputs!)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If we're drawing to a surface that's 2048 x 2048 pixels or larger there's
danger of fixed-point overflow in the triangle rasterization code. That
leads to various rendering glitches.
Rather than implement some intricate changes to the rasterization code,
simply subdivide triangles into smaller subtriangles to avoid the issue.
Only do this when the drawing surface is larger than 2048 by 2048.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Use the __builtin_ffs, __builtin_ffsll functions whenever we have GCC,
not just for specific platforms. Fixes Solaris build.
Note: This is a candidate for the stable branches.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=62868
Signed-off-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This can be enabled everywhere that ARB_texture_multisample is
supported -- ARB_texture_storage is supported on everything.
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Brian Paul <brianp@vmware.com>
Paul Berry
Roland Scheidegger
Vadim Girlin
Marek Olšák
Brian Paul
Christian König
Michel Dänzer
Maarten Lankhorst
Aras Pranckevicius
Eric Anholt
Vincent Lejeune
Haixia Shi
Matt Turner
Mike Lothian
Adam Jackson
Dave Airlie
Dave Airlie
Chris Forbes