In Vulkan, we'll compile the TCS and TES at the same time, so I can just
pass the TCS output VUE map to brw_compile_tes as the TES input VUE map.
So, we only need to do this in GL. Move it to the GL-specific layer.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This isn't needed, and Vulkan doesn't have one.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Moving this down a layer lets us share code between Vulkan and GL.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
NIR exists in both GL and Vulkan, but gl_program is GL specific.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
It feels weird using GL_* enums in a Vulkan driver.
v2: Fix the TESS_SPACING -> PIPE_TESS_SPACING conversion.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The vertex order is either clockwise or counterclockwise. We can just
store a "ccw" boolean rather than GLenum values. I don't want to use
GLenums in a Vulkan driver, and even in GL a simple boolean works fine.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
I apparently broke mark_whole_variable in ir_set_program_inouts.
It was passing a type that wasn't var->type, so the wrapper didn't
work out. It's all broken, revert it and start over.
Fixes all kinds of things on other drivers.
Revert "glsl: Make is_fixed_function_array actually check for varyings."
This reverts commit 42699e1271.
Revert "glsl: Mark whole variable used for ClipDistance and TessLevel*."
This reverts commit 5c580e64cc.
Revert "glsl: Override the # of varying slots for ClipDistance and TessLevel*."
This reverts commit 8b5749f65a.
Revert "glsl: Create and use a new ir_variable::count_attribute_slots() wrapper."
This reverts commit 6aa5cb34d0.
We can't check VARYING_SLOT_* locations until we've determined that
the variable is actually a varying.
Fixes assert failures in drivers which actually use this path,
such as radeonsi and i915.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99314
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
See also <https://bugs.freedesktop.org/show_bug.cgi?id=93551#c27> where
this was first observed as a requirement.
Signed-off-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Marking operations as redundant if they are equal to the base
range is fine when the tree structure is something like this:
max
/ \
max b
/ \
3 max
/ \
3 a
But the opt falls apart with a tree like this:
max
/ \
max max
/ \ / \
3 a b 3
The problem is that both branches are treated the same: descending in
the left branch will prune the constant, and then descending the right
branch will prune the constant there as well, because limits[0] wasn't
updated to take the change on the left branch into account, and so we
still get [3,\infty) as baserange.
In order to fix the bug we just disable the marking of redundant expressions
when they match the baserange.
NIR algebraic opt will clean up the first tree for anyway, hopefully
other backends are smart enough to do this also.
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
We run this after nir_lower_vars_to_ssa so that as many load/store_var
intrinsics as possible before copy_prop_vars executes. This is because the
pass isn't particularly efficient (it does a lot of linear walks of a
linked list) so we'd like as much of the work as possible to be done before
copy_prop_vars runs.
Shader DB results on Sky Lake:
total instructions in shared programs: 12020290 -> 12013627 (-0.06%)
instructions in affected programs: 26033 -> 19370 (-25.59%)
helped: 16
HURT: 13
total cycles in shared programs: 137772848 -> 137549012 (-0.16%)
cycles in affected programs: 6955660 -> 6731824 (-3.22%)
helped: 217
HURT: 237
total loops in shared programs: 3208 -> 3208 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0
total spills in shared programs: 4112 -> 4057 (-1.34%)
spills in affected programs: 483 -> 428 (-11.39%)
helped: 2
HURT: 0
total fills in shared programs: 5519 -> 5102 (-7.56%)
fills in affected programs: 993 -> 576 (-41.99%)
helped: 2
HURT: 0
LOST: 0
GAINED: 0
Broadwell had similar results. On older hardware, the impact isn't as
large because they don't advertise GL 4.5. Of the hurt programs, all but
one are hurt by a single instruction and the one is hurt by 3 instructions.
All of the helped programs, on the other hand, are helped by at least 3
instructions and one kerbal space program shader is helped by 44.59%.
The real star of the show, however, is the Gl43CSDof synmark2 benchmark
which has two shaders which are cut by 28% and 40% and the over-all runtime
performance of the benchmark on my Sky Lake laptop is improved by around
25-30% (it's a bit hard to be exact due to thermal throttling).
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Because border color is handled pre-swizzle, when we move the alpha
channel around in the format, the OPAQUE_BLACK border colors don't work
correctly on B4G4R4A4_UNORM_PACK16 with the hack. This fixes the
following Vulkan CTS tests on Broadwell:
dEQP-VK.pipeline.sampler.view_type.2d_array.format.b4g4r4a4_unorm_pack16.address_modes.all_mode_clamp_to_border_opaque_black
dEQP-VK.pipeline.sampler.view_type.1d_array.format.b4g4r4a4_unorm_pack16.address_modes.all_mode_clamp_to_border_opaque_black
dEQP-VK.pipeline.sampler.view_type.2d.format.b4g4r4a4_unorm_pack16.address_modes.all_mode_clamp_to_border_opaque_black
dEQP-VK.pipeline.sampler.view_type.1d.format.b4g4r4a4_unorm_pack16.address_modes.all_mode_clamp_to_border_opaque_black
dEQP-VK.pipeline.sampler.view_type.3d.format.b4g4r4a4_unorm_pack16.address_modes.all_mode_clamp_to_border_opaque_black
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
Interpreting layerCount literally would try to create billions of image
views in radv_process_depth_image_inplace().
Signed-off-by: Pierre-Loup A. Griffais <pgriffais@valvesoftware.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Treating everything as scalar arrays allows us to drop a bunch of
special case input/output munging all throughout the backend.
Instead, we just need to remap the TessLevel components to the
appropriate patch URB header locations in remap_patch_urb_offsets().
We also switch to treating the TES input versions of these as ordinary
shader inputs rather than system values, as remap_patch_urb_offsets()
just makes everything work out without special handling.
This regresses one Piglit test:
arb_tessellation_shader-large-uniforms/GL_TESS_CONTROL_SHADER-array-at-limit
The compiler starts promoting the constant arrays assigned to gl_TessLevel*
to uniform arrays. Since the shader also has a uniform array that uses
the maximum number of uniform components, this puts it over the uniform
component limit enforced by the linker. This is arguably a bug in the
constant array promotion code (it should avoid pushing us over limits),
but is unlikely to penalize any real application.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
It's only used in one place, it ignores the offset parameter currently,
and I want to add more parameters...at which point, passing in a bunch
of integers seems less obvious than writing it out.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This is harmless today because gl_TessLevelInner/Outer in the TES is
currently treated as system values. However, when we move to treating
them as inputs, this would cause a bug: with no TCS present, it would
propagate TES reads of VARYING_SLOT_TESS_LEVEL into the VS output VUE
map slots. This is totally bogus - those don't even exist in the VS.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Upcoming reworks in i965 are going to make it easy to handle this
like any other input. Having it as a system value will just require
additional code for no benefit.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
There's no point in trying to mark partial array access for
gl_ClipDistance, gl_TessLevelOuter, or gl_TessLevelInner - they're
special built-in variables that control fixed function hardware,
and will likely be used in an all-or-nothing fashion.
Since these arrays only occupy 1-2 varying slots, we have to avoid
our normal processing which increments the slot value by the array
index.
(I wrote this code before i965 switched from ir_set_program_inouts
to nir_shader_gather_info. It's not used by anyone today, and I'm
not sure how valuable it is...the alternative to GLSL IR lowering
is NIR compact arrays, at which point you should use nir_gather_info.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Right now, this shouldn't have any effect, as all drivers use
LowerClipDist and LowerTessFactors to turn the float[] arrays into
vectors.
However, it should help make it possible for drivers to avoid that
lowering.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
This wraps glsl_type::count_attribute_slots(), but will soon contain a
couple of overrides for a couple of GLSL built-ins variables.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Generally we should do tranpose after conversion, if the format has less than
32 bits per channel (if it has 32 bits, conversion is going to be a no-op
anyway...). This is obviously because there's less vectors to deal with.
Though the advantage for 16 bit formats isn't that big, and in fact with AVX
there isn't really any (as the 32bit unpacks can be done with 256bit, but
the smaller ones cannot, although that would change again with proper AVX2
support).
Only makes sense for 2d and not 1d cases. And to keep things easy, only handle
1,2 and 4 channels (rgbx is just fine).
For rgba unorm8 format the backend conversion sums up to these instruction
totals (not counting the movs for SSE2 due to 2-op syntax - generally every 2
unpacks need an additional mov).
SSE2 AVX
transpose: 32 unpack 16 unpack
untwiddle: 0 8 (128bit low/high permutes)
convert: 16 mul + 16 cvt 8 mul + 8 cvt
32->8bit: 12 pack 8 (128bit extract) + 12 pack
When doing transpose/untwiddle afterwards we get:
convert: 16 mul + 16 cvt 8 mul + 8 cvt
32->8bit: 12 pack 8 (128bit extract) + 12 pack
transpose/untwiddle 12 unpack 12 unpack
So for SSE2, this drops 20 unpacks (total instruction count 76->56)
whereas for AVX it replaces the 16 256bit unpacks with 8 128bit ones
and drops the 8 lo/hi permutes (in total 60->48). (Albeit to be fair,
the permutes could be dropped even when doing the transpose first,
they are extremely pointless but we'd need to be able to tell
lp_build_conv to reorder the vectors, for AVX2 we're going to need to
be able to tell lp_build_conv about ordering in any case.)
(With different ordering going into conversion, it would be possible
to do 4 unpacks + 4 pshufbs instead of 12 unpacks, but that might not
be better, and not all cpus can do it. Proper AVX2 support should eliminate
the 8 128bit extracts, reduce these 12 packs to 6 and the 12 unpacks to 2
pshufb + 2 permq ideally (+ 2 final 128bit extracts).)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This special packing path can be easily extended to handle not just
float->unorm8 but also float->snorm8 and uint32->uint8 and int32->int8
(i.e. all interesting cases for llvmpipe fs backend code).
The packing parts all stay the same (only the last step packing will
be signed->signed instead of signed->unsigned but luckily even sse2 can do
both).
While here also note some bugs with that (we keep the bugs identical to
what we did before on x86, albeit other archs may differ). In particular
float->unorm8 too large values will still get clamped to 0, not 255, and for
float->snorm8 NaNs will end up as -1, not 0 (but we do the clamp against 1.0
there to prevent too large values ending up as -1.0 - this is inconsistent
to unorm8 handling but is what we ended up before, I'm not sure we can get
away without it). This is quite fishy in any case as we depend on
arch-dependent behavior of the iround (my understanding is in fact with
altivec the conversion would actually saturate although I've no idea about
NaNs, so probably wouldn't need to do anything for snorm).
(There are only minimal piglit tests for unorm clamping behavior AFAICT, in
particular nothing seems to test values which are too large to be handled by
the float->int conversion.)
For uint32->uint8 we also do a min against MAX_INT, since the source for
the packs is always signed (again, on x86 - should probably be able to
express these arch-dependent bits better some day).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
For rgbx formats, there is no point in doing alpha conversion again (and
with different tranpose even, so llvm can't eliminate it).
Albeit it looks like there's some minimal changes needed in the blend code
(found by code inspection, no test seemed to complain) if we do this -
the blend factors are already sanitized if we have no destination alpha,
however for src_alpha_saturate it looks like it still might make a
difference (note that we forced has_alpha to true before for some formats
and nothing complained, but this seems safer).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
llvm has _huge_ problems trying to load things like <4 x i8> vectors and
stitching such loads together to form 128bit vectors. My understanding
of the problem is that the type legalizer tries to extend that to
really a <4 x i32> vector and not a <16 x i8> vector with the 4 elements
first then followed by padding, so the shuffles for then combining things
together are more or less impossible - you can in fact see the pmovzxd
llvm generates. Pre-4.0 llvm just gives up on it completely and does a 30+
pextrb/pinsrb sequence instead.
It looks like current llvm has fixed this behavior (my guess would be
due to better shuffle combination and load/shuffle folds), but we can
avoid this by just loading as <1 x i32> values, combine that and only
cast at the end. (I suspect it might also work if we'd pad the loaded
vectors immediately before shuffling them together, instead of directly
stitching 2 such vectors together pairwise before combining the pair.
But this _might_ lose the ability to load the values directly into
their right place in the vector with pinsrd.). But using 32bit values
is probably easier for llvm as it will never give it funny ideas how
the vector should look like.
(This is possibly only a problem for 1x8bit formats, since 2x8bit will
end up fetching 64bit hence only two vectors are stitched together,
not 4, but we use the same strategy anyway.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The only reason we didn't previously enable this was the dependency on
OpenGL ES 3.1. These should have been enabled as soon as HSW got
stencil texturing. We also needed to fixup setting MaxViewports.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Since 9d6ca7c3, there should be no performance hit for having
MaxViewports > 1. Always set this context state. This eliminates the
need to update this conditional as we add support for OES_viewport_array
on older GPUs.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The CS thread is needed to ensure proper ordering of operations and can't
be disabled (without complicating the code).
Discovered by Nine CSMT, which ended up in a deadlock.
Acked-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
it doesn't interact with compute shaders in any way
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
