mesa.git
5 years agogk110/ir: Use the new rcp/rsq in library
Boyan Ding [Thu, 9 Mar 2017 05:55:19 +0000 (13:55 +0800)]
gk110/ir: Use the new rcp/rsq in library

v2: (Karol Herbst <kherbst@redhat.com>
 * fix Value setup for the builtins

Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
[imirkin: track the fp64 flag when switching ops to calls]
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agogk110/ir: Add rsq f64 implementation
Boyan Ding [Thu, 9 Mar 2017 05:55:18 +0000 (13:55 +0800)]
gk110/ir: Add rsq f64 implementation

Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agogk110/ir: Add rcp f64 implementation
Boyan Ding [Thu, 9 Mar 2017 05:55:17 +0000 (13:55 +0800)]
gk110/ir: Add rcp f64 implementation

Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agonvc0: stick zero values for the compute invocation counts
Ilia Mirkin [Tue, 5 Feb 2019 08:05:33 +0000 (03:05 -0500)]
nvc0: stick zero values for the compute invocation counts

Not quite perfect, but at least we don't end up with random values in
the query buffer.

Fixes KHR-GL45.pipeline_statistics_query_tests_ARB.functional_default_qo_values

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agonv50,nvc0: use condition for occlusion queries when already complete
Ilia Mirkin [Tue, 5 Feb 2019 03:57:06 +0000 (22:57 -0500)]
nv50,nvc0: use condition for occlusion queries when already complete

For the NO_WAIT variants, we would jump into the ALWAYS case for both
nested and inverted occlusion queries. However if the query had
previously completed, the application could reasonably expect that the
render condition would follow that result.

To resolve this, we remove the nesting distinction which unnecessarily
created an imbalance between the regular and inverted cases (since
there's no "zero" condition mode). We also use the proper comparison if
we know that the query has completed (which could happen as a result of
an earlier get_query_result call).

Fixes KHR-GL45.conditional_render_inverted.functional

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agonvc0: fix 3d images on kepler
Ilia Mirkin [Mon, 4 Feb 2019 02:55:12 +0000 (21:55 -0500)]
nvc0: fix 3d images on kepler

Looks like SUBFM.3D and SUEAU are perfectly capable of dealing with 3d
tiling, they just need the correct inputs. Supply them.

We also have to deal with the case where a 2d "layer" of a 3d image is
bound. In this case, we supply the z coordinate separately to the
shader, which has to optionally treat every 2d case as if it could be a
slice of a 3d texture.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agonvc0/ir: fix second tex argument after levelZero optimization
Ilia Mirkin [Sat, 2 Feb 2019 07:56:48 +0000 (02:56 -0500)]
nvc0/ir: fix second tex argument after levelZero optimization

We used to pre-set a bunch of extra arguments to a texture instruction
in order to force the RA to allocate a register at the boundary of 4.
However with the levelZero optimization, which removes a LOD argument
when it's uniformly equal to zero, we undid that logic by removing an
extra argument. As a result, we could end up with insufficient alignment
on the second wide texture argument.

Instead we switch to a different method of achieving the same result.
The logic runs during the constraint analysis of the RA, and adds unset
sources as necessary right before being merged into a wide argument.

Fixes MISALIGNED_REG errors in Hitman when run with bindless textures
enabled on a GK208.

Fixes: 9145873b152 ("nvc0/ir: use levelZero flag when the lod is set to 0")
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agonvc0/ir: always use CG mode for loads from atomic-only buffers
Ilia Mirkin [Sun, 3 Feb 2019 15:06:24 +0000 (10:06 -0500)]
nvc0/ir: always use CG mode for loads from atomic-only buffers

Atomic operations don't update the local cache, which means that we
would have to issue CCTL operations in order to get the updated values.
When we know that a buffer is primarily used for atomic operations, it's
easier to just avoid the caching at that level entirely.

The same issue persists for non-atomic buffers, which will have to be
fixed separately.

Fixes the failing dEQP-GLES31.functional.atomic_counter.* tests.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agonvc0: add support for handling indirect draws with attrib conversion
Ilia Mirkin [Sat, 2 Feb 2019 21:23:50 +0000 (16:23 -0500)]
nvc0: add support for handling indirect draws with attrib conversion

The hardware does not natively support FIXED and DOUBLE formats. If
those are used in an indirect draw, they have to be converted. Our
conversion tries to be clever about only converting the data that's
needed. However for indirect, that won't work.

Given that DOUBLE or FIXED are highly unlikely to ever be used with
indirect draws, read the indirect buffer on the CPU and issue draws
directly.

Fixes the failing dEQP-GLES31.functional.draw_indirect.random.* tests.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agofreedreno/a6xx: Use tiling for all resources
Kristian H. Kristensen [Wed, 6 Feb 2019 22:10:36 +0000 (14:10 -0800)]
freedreno/a6xx: Use tiling for all resources

We used to restrict this to just PIPE_BIND_SAMPLER_VIEW resources, but
most resources benefit from being tiled.

Signed-off-by: Kristian H. Kristensen <hoegsberg@chromium.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>
5 years agofreedreno/a6xx: Emit blitter dst with OUT_RELOCW
Kristian H. Kristensen [Wed, 6 Feb 2019 21:47:32 +0000 (13:47 -0800)]
freedreno/a6xx: Emit blitter dst with OUT_RELOCW

We're writing to the bo and the kernel needs to know for
fd_bo_cpu_prep() to work.

Fixes: f93e43127252679b ("freedreno/a6xx: Enable blitter")
Reviewed-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Kristian H. Kristensen <hoegsberg@chromium.org>
5 years agoradv: Implement VK_EXT_buffer_device_address.
Bas Nieuwenhuizen [Thu, 24 Jan 2019 01:06:27 +0000 (02:06 +0100)]
radv: Implement VK_EXT_buffer_device_address.

v2: Also update the release notes.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoradv: Do not use the bo list for local buffers.
Bas Nieuwenhuizen [Fri, 25 Jan 2019 00:56:45 +0000 (01:56 +0100)]
radv: Do not use the bo list for local buffers.

The kernel already does it for us.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: Implement global memory accesses.
Bas Nieuwenhuizen [Thu, 24 Jan 2019 01:04:10 +0000 (02:04 +0100)]
amd/common: Implement global memory accesses.

Needed for VK_EXT_buffer_device_address.

The pointers are implmemented as i8*, since I could not figure
out how to emulate setting struct offsets in LLVM based on the
SPIR-V offsets (and more weird stuff like row major matrices).

Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: Do not use 32-bit loads for shared memory.
Bas Nieuwenhuizen [Thu, 24 Jan 2019 00:28:16 +0000 (01:28 +0100)]
amd/common: Do not use 32-bit loads for shared memory.

We use a straight glsl->llvm type conversion so types should already be right.

Also even though the writemasks were changed we we not actually doing 32-bit
things, so this fails miserably.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: handle nir_deref_cast for shared memory from integers.
Bas Nieuwenhuizen [Thu, 24 Jan 2019 00:25:50 +0000 (01:25 +0100)]
amd/common: handle nir_deref_cast for shared memory from integers.

Can happen e.g. after a phi.

Fixes: a2b5cc3c399 "radv: enable variable pointers"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: Handle nir_deref_type_ptr_as_array for shared memory.
Bas Nieuwenhuizen [Thu, 24 Jan 2019 00:21:28 +0000 (01:21 +0100)]
amd/common: Handle nir_deref_type_ptr_as_array for shared memory.

Fixes: a2b5cc3c399 "radv: enable variable pointers"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: Fix stores to derefs with unknown variable.
Bas Nieuwenhuizen [Wed, 23 Jan 2019 21:50:33 +0000 (22:50 +0100)]
amd/common: Fix stores to derefs with unknown variable.

Fixes: a2b5cc3c399 "radv: enable variable pointers"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: Use correct writemask for shared memory stores.
Bas Nieuwenhuizen [Wed, 23 Jan 2019 21:44:05 +0000 (22:44 +0100)]
amd/common: Use correct writemask for shared memory stores.

The check was for 1 bit being set, which is clearly not what we want.

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoradv: Fix the shader info pass for not having the variable.
Bas Nieuwenhuizen [Wed, 23 Jan 2019 21:41:46 +0000 (22:41 +0100)]
radv:  Fix the shader info pass for not having the variable.

For example with VK_EXT_buffer_device_address or
 VK_KHR_variable_pointers.

Fixes: a2b5cc3c399 "radv: enable variable pointers"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: Implement ptr->int casts in ac_to_integer.
Bas Nieuwenhuizen [Wed, 23 Jan 2019 00:54:39 +0000 (01:54 +0100)]
amd/common: Implement ptr->int casts in ac_to_integer.

For the implicit casts inherent in nir.

This should probably have been done for shared memory for
VK_KHR_variable_pointers.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoamd/common: Add gep helper for pointer increment.
Bas Nieuwenhuizen [Wed, 23 Jan 2019 00:53:59 +0000 (01:53 +0100)]
amd/common: Add gep helper for pointer increment.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoradv: Only look at pImmutableSamples if the descriptor has a sampler.
Bas Nieuwenhuizen [Tue, 22 Jan 2019 21:29:26 +0000 (22:29 +0100)]
radv: Only look at pImmutableSamples if the descriptor has a sampler.

Equivalent of ANV patch c7f4a2867ce492d78c1f8e2870c0a593d280572d

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoxvmc: fix string comparison
Eric Engestrom [Wed, 6 Feb 2019 16:28:12 +0000 (16:28 +0000)]
xvmc: fix string comparison

Fixes: 6fca18696d0e6a243f6f "g3dvl: Update XvMC unit tests."
Cc: Younes Manton <younes.m@gmail.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agoxvmc: fix string comparison
Eric Engestrom [Wed, 6 Feb 2019 16:21:08 +0000 (16:21 +0000)]
xvmc: fix string comparison

Fixes: c7b65dcaffeb9d0760c8 "xvmc: Define some Xv attribs to allow users
                             to specify color standard and procamp"
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson glvnd build
Eric Engestrom [Tue, 29 Jan 2019 08:57:17 +0000 (08:57 +0000)]
gitlab-ci: add meson glvnd build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agotravis: remove unused scons code path
Eric Engestrom [Mon, 28 Jan 2019 21:36:19 +0000 (21:36 +0000)]
travis: remove unused scons code path

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agotravis: remove unused linux code path
Eric Engestrom [Mon, 28 Jan 2019 21:36:08 +0000 (21:36 +0000)]
travis: remove unused linux code path

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium ST Other build
Eric Engestrom [Mon, 28 Jan 2019 20:48:37 +0000 (20:48 +0000)]
gitlab-ci: add make Gallium ST Other build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium ST Clover LLVM-7 build
Eric Engestrom [Mon, 28 Jan 2019 20:40:25 +0000 (20:40 +0000)]
gitlab-ci: add make Gallium ST Clover LLVM-7 build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium ST Clover LLVM-6.0 build
Eric Engestrom [Mon, 28 Jan 2019 20:39:22 +0000 (20:39 +0000)]
gitlab-ci: add make Gallium ST Clover LLVM-6.0 build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium ST Clover LLVM-5.0 build
Eric Engestrom [Mon, 28 Jan 2019 20:37:03 +0000 (20:37 +0000)]
gitlab-ci: add make Gallium ST Clover LLVM-5.0 build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium ST Clover LLVM-4.0 build
Eric Engestrom [Mon, 28 Jan 2019 20:36:25 +0000 (20:36 +0000)]
gitlab-ci: add make Gallium ST Clover LLVM-4.0 build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium ST Clover LLVM-3.9 build
Eric Engestrom [Mon, 28 Jan 2019 20:29:59 +0000 (20:29 +0000)]
gitlab-ci: add make Gallium ST Clover LLVM-3.9 build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium Drivers "Other" build
Eric Engestrom [Mon, 28 Jan 2019 20:24:04 +0000 (20:24 +0000)]
gitlab-ci: add make Gallium Drivers "Other" build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium Drivers RadeonSI build
Eric Engestrom [Mon, 28 Jan 2019 20:21:44 +0000 (20:21 +0000)]
gitlab-ci: add make Gallium Drivers RadeonSI build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make Gallium Drivers SWR build
Eric Engestrom [Mon, 28 Jan 2019 20:17:12 +0000 (20:17 +0000)]
gitlab-ci: add make Gallium Drivers SWR build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add make loaders/classic DRI build
Eric Engestrom [Mon, 28 Jan 2019 20:13:33 +0000 (20:13 +0000)]
gitlab-ci: add make loaders/classic DRI build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson gallium ST "Other" build
Eric Engestrom [Mon, 28 Jan 2019 18:13:55 +0000 (18:13 +0000)]
gitlab-ci: add meson gallium ST "Other" build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson gallium ST Clover (LLVM 7.0) build
Eric Engestrom [Mon, 28 Jan 2019 18:09:24 +0000 (18:09 +0000)]
gitlab-ci: add meson gallium ST Clover (LLVM 7.0) build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson gallium ST Clover (LLVM 6.0) build
Eric Engestrom [Mon, 28 Jan 2019 18:07:31 +0000 (18:07 +0000)]
gitlab-ci: add meson gallium ST Clover (LLVM 6.0) build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson gallium ST Clover (LLVM 5.0) build
Eric Engestrom [Mon, 28 Jan 2019 18:05:22 +0000 (18:05 +0000)]
gitlab-ci: add meson gallium ST Clover (LLVM 5.0) build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson gallium "other drivers" build
Eric Engestrom [Mon, 28 Jan 2019 17:47:09 +0000 (17:47 +0000)]
gitlab-ci: add meson gallium "other drivers" build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson gallium RadeonSI build
Eric Engestrom [Mon, 28 Jan 2019 17:45:26 +0000 (17:45 +0000)]
gitlab-ci: add meson gallium RadeonSI build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson gallium SWR build
Eric Engestrom [Mon, 28 Jan 2019 17:42:59 +0000 (17:42 +0000)]
gitlab-ci: add meson gallium SWR build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add meson loader/classic DRI build
Eric Engestrom [Mon, 28 Jan 2019 17:38:17 +0000 (17:38 +0000)]
gitlab-ci: add meson loader/classic DRI build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add scons SWR build
Eric Engestrom [Mon, 28 Jan 2019 16:33:22 +0000 (16:33 +0000)]
gitlab-ci: add scons SWR build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add scons llvm 3.5 build
Eric Engestrom [Mon, 28 Jan 2019 16:30:36 +0000 (16:30 +0000)]
gitlab-ci: add scons llvm 3.5 build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add a scons no-llvm build
Eric Engestrom [Wed, 23 Jan 2019 15:46:10 +0000 (15:46 +0000)]
gitlab-ci: add a scons no-llvm build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add a make vulkan build
Eric Engestrom [Mon, 21 Jan 2019 09:42:37 +0000 (09:42 +0000)]
gitlab-ci: add a make vulkan build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add a meson vulkan build
Eric Engestrom [Sun, 20 Jan 2019 11:26:53 +0000 (11:26 +0000)]
gitlab-ci: add a meson vulkan build

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agogitlab-ci: add ubuntu container
Eric Engestrom [Sun, 20 Jan 2019 11:21:45 +0000 (11:21 +0000)]
gitlab-ci: add ubuntu container

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
5 years agoradeonsi: use local ws variable in si_need_dma_space
Marek Olšák [Thu, 31 Jan 2019 15:59:48 +0000 (10:59 -0500)]
radeonsi: use local ws variable in si_need_dma_space

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
5 years agoradeonsi: don't leak an index buffer if draw_vbo fails
Marek Olšák [Tue, 22 Jan 2019 22:29:19 +0000 (17:29 -0500)]
radeonsi: don't leak an index buffer if draw_vbo fails

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
5 years agoradeonsi: make allocator_zeroed_memory unmappable and use bigger buffers
Marek Olšák [Wed, 30 Jan 2019 19:29:36 +0000 (14:29 -0500)]
radeonsi: make allocator_zeroed_memory unmappable and use bigger buffers

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
5 years agoradeonsi: clear allocator_zeroed_memory with SDMA
Marek Olšák [Tue, 29 Jan 2019 00:49:56 +0000 (19:49 -0500)]
radeonsi: clear allocator_zeroed_memory with SDMA

so that it can be used in parallel IBs.

This also removes the SO_FILLED_SIZE hack.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
5 years agoradeonsi: initialize textures using DCC to black when possible
Marek Olšák [Wed, 23 Jan 2019 16:34:48 +0000 (11:34 -0500)]
radeonsi: initialize textures using DCC to black when possible

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
5 years agofreedreno: a2xx: fix fast clear
Jonathan Marek [Tue, 5 Feb 2019 19:08:12 +0000 (14:08 -0500)]
freedreno: a2xx: fix fast clear

Fixes: 912a9c8d
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agoegl: use coherent variable names
Eric Engestrom [Sat, 2 Feb 2019 11:38:45 +0000 (11:38 +0000)]
egl: use coherent variable names

`EGLDisplay` variables (the opaque Khronos type) have mostly been
consistently called `dpy`, as this is the name used in the Khronos
specs.

However, `_EGLDisplay` variables (our internal struct) have been
randomly called `dpy` when there was no local variable clash with
`EGLDisplay`s, and `disp` otherwise.

Let's be consistent and use `dpy` for the Khronos type, and `disp`
for our struct.

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Eric Anholt <eric@anholt.net>
5 years agomeson: Remove panfrost from default driver list
Alyssa Rosenzweig [Tue, 5 Feb 2019 02:00:12 +0000 (02:00 +0000)]
meson: Remove panfrost from default driver list

Until the kernel side matures and the full driver is upstreamed, to
avoid end-user surprises, Panfrost should only be built for the
adventurous.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agov3d: Whitespace consistency fix.
Eric Anholt [Mon, 4 Feb 2019 20:07:41 +0000 (12:07 -0800)]
v3d: Whitespace consistency fix.

5 years agov3d: Fix copy-propagation of input unpacks.
Eric Anholt [Mon, 4 Feb 2019 18:35:40 +0000 (10:35 -0800)]
v3d: Fix copy-propagation of input unpacks.

I had a single function for "does this do float input unpacking" with two
major flaws: It was missing the most common thing to try to copy propagate
a f32 input nunpack to (the VFPACK to an FP16 render target) along with
several other ALU ops, and also would try to propagate an f32 unpack into
a VFMUL which only does f16 unpacks.

instructions in affected programs: 659232 -> 655895 (-0.51%)
uniforms in affected programs: 132613 -> 135336 (2.05%)

and a couple of programs increase their thread counts.

The uniforms hit appears to be a pattern in generated code of doing (-a >=
a) comparisons, which when a is abs(b) can result in the abs instruction
being copy propagated once but not fully DCEed.

5 years agov3d: Fix input packing of .l for rounding/fdx/fdy.
Eric Anholt [Tue, 5 Feb 2019 01:15:36 +0000 (17:15 -0800)]
v3d: Fix input packing of .l for rounding/fdx/fdy.

Avoids a regression in
dEQP-GLES3.functional.shaders.derivate.fwidth.texture.* once we start
copy-propagating more input packs.

5 years agov3d: Fix pack/unpack of VFPACK operand unpacks.
Eric Anholt [Mon, 4 Feb 2019 19:18:55 +0000 (11:18 -0800)]
v3d: Fix pack/unpack of VFPACK operand unpacks.

We want to be able to copy propagate our texture unpacks into the vfpack.

5 years agov3d: Fix dumping of shaders with alpha test.
Eric Anholt [Sat, 2 Feb 2019 01:18:41 +0000 (17:18 -0800)]
v3d: Fix dumping of shaders with alpha test.

We were trying to print a NULL entry from the table.

5 years agov3d: Store the actual mask of color buffers present in the key.
Eric Anholt [Fri, 1 Feb 2019 21:13:48 +0000 (13:13 -0800)]
v3d: Store the actual mask of color buffers present in the key.

If you only bound rt 1+, we'd still emit a write to the rt0 that isn't
present (noticed while debugging an
ext_framebuffer_multisample-alpha-to-coverage-no-draw-buffer-zero
regression in another change).

5 years agov3d: Fix precompile of FRAG_RESULT_DATA1 and higher outputs.
Eric Anholt [Sat, 2 Feb 2019 00:41:56 +0000 (16:41 -0800)]
v3d: Fix precompile of FRAG_RESULT_DATA1 and higher outputs.

I was just leaving the other MRT targets than DATA0 out, by accident.

5 years agost/nir: Use src/ relative include path for autotools
Kristian H. Kristensen [Tue, 5 Feb 2019 22:13:51 +0000 (14:13 -0800)]
st/nir: Use src/ relative include path for autotools

Fixes: cdc53fa81cbeb80373eac33ef7695d9025caf14b
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Kristian H. Kristensen <hoegsberg@chromium.org>
5 years agogallium: Add a PIPE_CAP_NIR_COMPACT_ARRAYS capability bit.
Kenneth Graunke [Wed, 23 Jan 2019 10:04:01 +0000 (02:04 -0800)]
gallium: Add a PIPE_CAP_NIR_COMPACT_ARRAYS capability bit.

Iris would like to use compact arrays for tesslevels and clip/cull
distances.  radeonsi will likely want to switch to these at some point,
since it'll be necessary for GL_ARB_gl_spirv support, but it's not ready
for them just yet.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agost/nir: Call nir_lower_clip_cull_distance_arrays().
Kenneth Graunke [Wed, 23 Jan 2019 10:44:28 +0000 (02:44 -0800)]
st/nir: Call nir_lower_clip_cull_distance_arrays().

Today, st always sets LowerCombinedClipCullDistance, causing the GLSL IR
lowering to run, giving us vec4[2] arrays.  I would like to disable this
and instead run the NIR lowering so that we get compact float[] arrays
instead.

Calling the new pass is a noop if the GLSL IR pass has already run, so
it's safe to call the pass unconditionally.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agonir: Avoid splitting compact arrays into per-element variables.
Kenneth Graunke [Thu, 5 Jul 2018 21:34:27 +0000 (14:34 -0700)]
nir: Avoid splitting compact arrays into per-element variables.

Compact arrays are used for special variables like clip and cull
distances, or tessellation levels.  Drivers using compact arrays
assume that these values will always be actual arrays.  We don't
want to turn a float[1] gl_CullDistance into a single float; that
would confuse drivers.

Today, i965 uses compact arrays, and Gallium drivers use
nir_lower_io_arrays_to_elements, so we haven't had any overlap
that would demonstrate the issue.  Iris will use both.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agonir: Avoid clip/cull distance lowering multiple times.
Kenneth Graunke [Wed, 23 Jan 2019 09:15:05 +0000 (01:15 -0800)]
nir: Avoid clip/cull distance lowering multiple times.

A couple places in st/nir assume that cull distances have been lowered
away, so it will need to call this lowering pass for drivers which opt
out of the GLSL IR lowering.  The Intel backend also calls this pass,
for i965 and anv.  We need to only do it once.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agonir: Bail on clip/cull distance lowering if GLSL IR already did it.
Kenneth Graunke [Wed, 23 Jan 2019 10:36:44 +0000 (02:36 -0800)]
nir: Bail on clip/cull distance lowering if GLSL IR already did it.

We have a GLSL IR pass to convert clip/cull distance float[] arrays
into vec4[2] arrays.  In ff281e6204, we attempted to skip this pass
if the GLSL IR lowering had already run.  But, that code was not quite
right, as we forgot to strip away the per-vertex IO array layer for
geometry and tessellation shader varyings.

If the GLSL IR pass has run, the variables will not be marked as
"compact".  So we can simply check that and bail.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agocompiler: Mark clip/cull distance arrays as compact before lowering.
Kenneth Graunke [Wed, 23 Jan 2019 09:55:45 +0000 (01:55 -0800)]
compiler: Mark clip/cull distance arrays as compact before lowering.

nir_lower_clip_cull_distance_arrays() marks the combined clip/cull
distance array as compact.  However, when translating in from GLSL
or SPIR-V, we were not marking the original float[] arrays as compact.

We should do so.  That way, we can detect these corner cases properly.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agonir: Record info->fs.pixel_center_integer in lower_system_values
Kenneth Graunke [Sat, 2 Feb 2019 08:43:42 +0000 (00:43 -0800)]
nir: Record info->fs.pixel_center_integer in lower_system_values

radeonsi uses a system value for gl_FragCoord rather than an input var.
These get translated into load_frag_coord NIR intrinsics, which lose the
pixel_center_integer and origin_upper_left decorations.  To cope with
this, Tim added a shader_info field for pixel_center_integer, and made
glsl_to_nir set it accordingly.

prog_to_nir also needs to handle these fragcoord conventions.  Instead
of duplicating the logic to set the info field, just move it to
nir_lower_system_values so it'll happen regardless of who makes the NIR.

(For what it's worth, we don't need an info flag for origin_upper_left,
because radeonsi lowers origin conventions in nir_lower_wpos_ytransform
before nir_lower_system_values destroys the variable and qualifiers.)

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agoprogram: Extend prog_to_nir handle system values.
Kenneth Graunke [Fri, 1 Feb 2019 05:52:50 +0000 (21:52 -0800)]
program: Extend prog_to_nir handle system values.

Some drivers, such as radeonsi, use a system value for gl_FragCoord
rather than an input variable.  In this case, our Mesa IR will have
a PROGRAM_SYSTEM_VALUE register, which we need to translate.

This makes prog_to_nir work for Gallium drivers which expose the
PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL capability bit.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agoprogram: Use u_bit_scan64 in prog_to_nir.
Kenneth Graunke [Fri, 1 Feb 2019 06:02:55 +0000 (22:02 -0800)]
program: Use u_bit_scan64 in prog_to_nir.

We can simply iterate the bits rather than using util_last_bit and
checking each one up until that point.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agost/mesa: Add NIR versions of the PBO upload/download shaders.
Kenneth Graunke [Fri, 2 Nov 2018 09:39:58 +0000 (02:39 -0700)]
st/mesa: Add NIR versions of the PBO upload/download shaders.

Acked-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Tested-by: Eric Anholt <eric@anholt.net>
5 years agost/mesa: Add a NIR version of the OES_draw_texture built-in shaders.
Kenneth Graunke [Sun, 4 Nov 2018 23:42:02 +0000 (15:42 -0800)]
st/mesa: Add a NIR version of the OES_draw_texture built-in shaders.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Tested-by: Eric Anholt <eric@anholt.net>
5 years agost/mesa: Add NIR versions of the clear shaders.
Kenneth Graunke [Fri, 2 Nov 2018 09:24:00 +0000 (02:24 -0700)]
st/mesa: Add NIR versions of the clear shaders.

We implement the basic VS and FS, as well as the VS that does layered
clears by writing gl_Layer from the vertex shader.  Drivers which need
a geometry shader for writing layer continue falling back to TGSI, as
I didn't need this and so didn't bother implementing it.  (We certainly
could, however, if people want to add it in the future.)

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Tested-by: Eric Anholt <eric@anholt.net>
5 years agost/mesa: Add NIR versions of the drawpixels Z/stencil fragment shaders.
Kenneth Graunke [Mon, 17 Sep 2018 02:08:07 +0000 (19:08 -0700)]
st/mesa: Add NIR versions of the drawpixels Z/stencil fragment shaders.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Tested-by: Eric Anholt <eric@anholt.net>
5 years agost/mesa: Add a NIR version of the drawpixels/bitmap VS copy shader.
Kenneth Graunke [Sun, 4 Nov 2018 06:46:29 +0000 (23:46 -0700)]
st/mesa: Add a NIR version of the drawpixels/bitmap VS copy shader.

This provides a native NIR version of the DrawPixels/Bitmap passthrough
vertex shader.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Tested-by: Eric Anholt <eric@anholt.net>
5 years agost/nir: Make new helpers for constructing built-in NIR shaders.
Kenneth Graunke [Tue, 30 Oct 2018 05:30:06 +0000 (22:30 -0700)]
st/nir: Make new helpers for constructing built-in NIR shaders.

The state tracker generates several built-in shaders in order to
perform scissored clears, upload/download PBOs, and so on.  These
are currently constructed using TGSI, using ureg and u_simple_shader.

I want to have NIR versions of these shaders, for my Gallium driver
that has a NIR backend but no TGSI support.  To that end, we'll want
a few helpers to help construct simple shaders.

This patch adds two new helpers:

- st_nir_finish_builtin_shader() takes a manually constructed NIR
  shader, applies lowering passes (like st_link_nir would do for GLSL),
  and constructs the pipe_shader_state.

- st_nir_make_passthrough_shader() makes a simple passthrough shader,
  which copies inputs to outputs.  This is similar to u_simple_shaders.

v2: Set info->fs.untyped_color_outputs for vc4/v3d (thanks Eric!).

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Tested-by: Eric Anholt <eric@anholt.net>
5 years agost/nir: Move varying setup code to a helper function.
Kenneth Graunke [Tue, 30 Oct 2018 05:41:18 +0000 (22:41 -0700)]
st/nir: Move varying setup code to a helper function.

I want to reuse this for built-in shaders.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Tested-by: Eric Anholt <eric@anholt.net>
5 years agonir/deref: Drop zero ptr_as_array derefs
Jason Ekstrand [Sun, 3 Feb 2019 15:32:32 +0000 (09:32 -0600)]
nir/deref: Drop zero ptr_as_array derefs

They are effectively (&x)[0] or *&x which does nothing.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agonir: Move V3D's "the shader was TGSI, ignore FS output types" flag to NIR.
Eric Anholt [Tue, 5 Feb 2019 18:22:30 +0000 (10:22 -0800)]
nir: Move V3D's "the shader was TGSI, ignore FS output types" flag to NIR.

Ken's rework of mesa/st builtins to NIR means that we'll have more NIR
shaders with color output types that are mismatched with the render target
types.  Since this is behavior that GLSL doesn't require, add it as a
shader_info option so the driver can know that it needs to ignore the FS
output's base type in favor of the actual render target's.  This prevents
needing additional variants in several mesa/st paths (clear, pbo upload,
pbo download), given that the driver already has to handle the variants
for any TGSI being passed to it (from u_blitter, for example).

Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
5 years agoanv: wire up the state_pool_padding test
Emil Velikov [Tue, 5 Feb 2019 12:09:45 +0000 (12:09 +0000)]
anv: wire up the state_pool_padding test

Cc: Jason Ekstrand <jason@jlekstrand.net>
Fixes: 927ba12b53c ("anv/tests: Adding test for the state_pool padding.")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com><Paste>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
5 years agonvc0/ir: replace cvt instructions with add to improve shader performance
Karol Herbst [Thu, 13 Dec 2018 19:19:45 +0000 (20:19 +0100)]
nvc0/ir: replace cvt instructions with add to improve shader performance

gives me an performance boost of 0.2% in pixmark_piano on my gk106, gm204 and
gp107.

reduces the amount of generated convert instructions by roughly 30% in
shader-db.

v2: only for 32 bit operations
    move some common code out of the switch
    handle OP_SAT with modifiers
v3: only for registers and const memory
    rework if clauses
    merge isCvt into this patch
v4: merge isCvt into its use

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
5 years agogallium-xlib: query MIT-SHM before using it.
Bart Oldeman [Sun, 3 Feb 2019 02:57:10 +0000 (02:57 +0000)]
gallium-xlib: query MIT-SHM before using it.

When Mesa is compiled for gallium-xlib using e.g.
./configure --enable-glx=gallium-xlib --disable-dri --disable-gbm
-disable-egl
and is used by an X server (usually remotely via SSH X11 forwarding)
that does not support MIT-SHM such as XMing or MobaXterm, OpenGL
clients report error messages such as
Xlib:  extension "MIT-SHM" missing on display "localhost:11.0".
ad infinitum.

The reason is that the code in src/gallium/winsys/sw/xlib uses
MIT-SHM without checking for its existence, unlike the code
in src/glx/drisw_glx.c and src/mesa/drivers/x11/xm_api.c.
I copied the same check using XQueryExtension, and tested with
glxgears on MobaXterm.

This issue was reported before here:
https://lists.freedesktop.org/archives/mesa-users/2016-July/001183.html

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Cc: <mesa-stable@lists.freedesktop.org>
5 years agoswr/rast: update SWR rasterizer shader stats
Alok Hota [Thu, 7 Jun 2018 18:14:48 +0000 (13:14 -0500)]
swr/rast: update SWR rasterizer shader stats

Primarily refactoring internal stats types

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
5 years agoloader/dri3: Use strlen instead of sizeof for creating VRR property atom
Michel Dänzer [Mon, 4 Feb 2019 17:53:52 +0000 (18:53 +0100)]
loader/dri3: Use strlen instead of sizeof for creating VRR property atom

sizeof counts the terminating null character as well, so that also
contributed to the ID computed for the X11 atom. But the convention is
for only the non-null characters to contribute to the atom ID.

Fixes: 2e12fe425fe3 "loader/dri3: Enable adaptive_sync via
                     _VARIABLE_REFRESH property"
Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir: add missing vec opcodes in lower_bool_to_float
Jonathan Marek [Tue, 29 Jan 2019 14:09:07 +0000 (09:09 -0500)]
nir: add missing vec opcodes in lower_bool_to_float

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agomesa: release references to image textures when a context is destroyed
Gert Wollny [Sat, 2 Feb 2019 12:17:16 +0000 (13:17 +0100)]
mesa: release references to image textures when a context is destroyed

When a texture is still bound as an image and the context it was bound in
is destroyed but not the texture, then the texture will still hold the
resource and will not be freed when it is finally destroyed. Hence, release
these references when the context is destroyed.

This leak was triggered by virglrenderer:
https://gitlab.freedesktop.org/virgl/virglrenderer/issues/86

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
5 years agoradeonsi: release tokens after creating the shader program
Gert Wollny [Thu, 31 Jan 2019 13:50:41 +0000 (14:50 +0100)]
radeonsi: release tokens after creating the shader program

ureg_get_tokens clears the reference to the tokens, and create_compute_state makes
a copy, hence the tokens must be explicitely released.

Fixes: Direct leak of 256 byte(s) in 1 object(s) allocated from:
    #0 0x7ff729cf3c60 in realloc (/usr/lib64/gcc/x86_64-pc-linux-gnu/7.3.0/libasan.so+0xdbc60)
    #1 0x7ff721b1240c in tokens_expand ../../samba/mesa/src/gallium/auxiliary/tgsi/tgsi_ureg.c:234
    #2 0x7ff721b1c9c0 in get_tokens ../../samba/mesa/src/gallium/auxiliary/tgsi/tgsi_ureg.c:257
    #3 0x7ff721b1c9c0 in copy_instructions ../../samba/mesa/src/gallium/auxiliary/tgsi/tgsi_ureg.c:2040
    #4 0x7ff721b1c9c0 in ureg_finalize ../../samba/mesa/src/gallium/auxiliary/tgsi/tgsi_ureg.c:2090
    #5 0x7ff721b1e919 in ureg_get_tokens ../../samba/mesa/src/gallium/auxiliary/tgsi/tgsi_ureg.c:2167
    #6 0x7ff721f8b35a in si_create_dma_compute_shader ../../samba/mesa/src/gallium/drivers/radeonsi/si_shaderlib_tgsi.c:219
    #7 0x7ff722043ed9 in si_compute_do_clear_or_copy ../../samba/mesa/src/gallium/drivers/radeonsi/si_compute_blit.c:156
    #8 0x7ff7220448d3 in si_clear_buffer ../../samba/mesa/src/gallium/drivers/radeonsi/si_compute_blit.c:247
    #9 0x7ff7220350e8 in vi_dcc_clear_level ../../samba/mesa/src/gallium/drivers/radeonsi/si_clear.c:274

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
5 years agoisl: assert that Gen8+ don't have bit6_swizzling
Caio Marcelo de Oliveira Filho [Thu, 31 Jan 2019 21:18:19 +0000 (13:18 -0800)]
isl: assert that Gen8+ don't have bit6_swizzling

v2: Rewrite the condition to more clearly match the comment. (Jordan)

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoanv: skip bit6 swizzle detection in Gen8+
Caio Marcelo de Oliveira Filho [Thu, 31 Jan 2019 21:29:40 +0000 (13:29 -0800)]
anv: skip bit6 swizzle detection in Gen8+

It is always false on Gen8+.  Also, move the variable definition near
its use.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoi965: skip bit6 swizzle detection in Gen8+
Caio Marcelo de Oliveira Filho [Thu, 31 Jan 2019 21:28:24 +0000 (13:28 -0800)]
i965: skip bit6 swizzle detection in Gen8+

It is always false on Gen8+.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agonir: keep the phi order when splitting blocks
Caio Marcelo de Oliveira Filho [Sat, 26 Jan 2019 09:05:14 +0000 (01:05 -0800)]
nir: keep the phi order when splitting blocks

All things being equal is better to keep the original order.  Since
the new block is empty, push the phis in order to tail.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Schürmann <daniel.schuermann@campus.tu-berlin.de>
5 years agonv50,nvc0: add explicit settings for recent caps
Ilia Mirkin [Tue, 5 Feb 2019 04:33:09 +0000 (23:33 -0500)]
nv50,nvc0: add explicit settings for recent caps

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
5 years agopanfrost: Implement Midgard shader toolchain
Alyssa Rosenzweig [Wed, 30 Jan 2019 01:11:31 +0000 (01:11 +0000)]
panfrost: Implement Midgard shader toolchain

This patch implements the free Midgard shader toolchain: the assembler,
the disassembler, and the NIR-based compiler. The assembler is a
standalone inaccessible Python script for reference purposes. The
disassembler and the compiler are implemented in C, accessible via the
standalone `midgard_compiler` binary. Later patches will use these
interfaces from the driver for online compilation.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Acked-by: Rob Clark <robdclark@gmail.com>
Acked-by: Eric Anholt <eric@anholt.net>
Acked-by: Emil Velikov <emil.velikov@collabora.com>