mesa.git
5 years agoradv/gfx10: set BREAK_WAVE_AT_EOI if TES or GS enable the primitive ID
Samuel Pitoiset [Thu, 18 Jul 2019 08:14:09 +0000 (10:14 +0200)]
radv/gfx10: set BREAK_WAVE_AT_EOI if TES or GS enable the primitive ID

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoradv/gfx10: move emitting VGT_PRIMITIVEID_EN into the NGG path
Samuel Pitoiset [Thu, 18 Jul 2019 08:14:07 +0000 (10:14 +0200)]
radv/gfx10: move emitting VGT_PRIMITIVEID_EN into the NGG path

And do not emit VGT_GS_MODE which is unnecessary on GFX10.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agoradv/gfx10: do not always execute a barrier before the second shader
Samuel Pitoiset [Wed, 17 Jul 2019 13:43:53 +0000 (15:43 +0200)]
radv/gfx10: do not always execute a barrier before the second shader

With NGG, empty waves may still be required to export data.

This fixes dEQP-VK.ycbcr.format.*_unorm.geometry_*.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agoradv: fix VGT_GS_MODE if VS uses the primitive ID
Samuel Pitoiset [Wed, 17 Jul 2019 08:58:17 +0000 (10:58 +0200)]
radv: fix VGT_GS_MODE if VS uses the primitive ID

Found by inspection.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agov3d: emit correct lowering for logic operations with MSAA render targets
Iago Toral Quiroga [Tue, 16 Jul 2019 08:58:55 +0000 (10:58 +0200)]
v3d: emit correct lowering for logic operations with MSAA render targets

v2:
 - Drop the writemask from the per-sample color intrinsic (Eric)

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agov3d: handle nir_intrinsic_store_tlb_sample_color_v3d
Iago Toral Quiroga [Tue, 16 Jul 2019 09:20:11 +0000 (11:20 +0200)]
v3d: handle nir_intrinsic_store_tlb_sample_color_v3d

v2:
 - Move handling of output intrinsics to ntq_emit_intrinsic() (Eric).

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir: add a V3D-specific intrinsic for per-sample color writes
Iago Toral Quiroga [Tue, 16 Jul 2019 07:44:52 +0000 (09:44 +0200)]
nir: add a V3D-specific intrinsic for per-sample color writes

For per-sample color writes we need the output intrinsic to pack the
sample index, which is not provided with regular store_output intrinsics
unless we figured out a way to encode it into the base or the offset.

v2:
 - Drop the writemask (Eric)

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agov3d: implement per-sample tlb color writes
Iago Toral Quiroga [Tue, 16 Jul 2019 08:48:06 +0000 (10:48 +0200)]
v3d: implement per-sample tlb color writes

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agov3d: refactor the tlb color write code
Iago Toral Quiroga [Tue, 16 Jul 2019 08:28:53 +0000 (10:28 +0200)]
v3d: refactor the tlb color write code

We want to split the tlb specifier setup from the color writes, because when
we implement per-sample color writes we want to do the latter for all the
samples, but the former only once.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agov3d: move tlb color write emission to a helper function
Iago Toral Quiroga [Tue, 16 Jul 2019 08:14:17 +0000 (10:14 +0200)]
v3d: move tlb color write emission to a helper function

We will soon be adding per-sample color writes which means additional
complexity and more indentation (we will need another loop to emit
the writes for each individual sample), so this will help keeping
things simple and a bit more readable.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agov3d: implement per-sample tlb color reads
Iago Toral Quiroga [Tue, 16 Jul 2019 08:58:12 +0000 (10:58 +0200)]
v3d: implement per-sample tlb color reads

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agoanv: fix format mapping for depth/stencil formats
Lionel Landwerlin [Wed, 17 Jul 2019 22:00:11 +0000 (01:00 +0300)]
anv: fix format mapping for depth/stencil formats

anv_format is supposed to have a pointer back to the associated
VkFormat, we were missed this for depth/stencil formats.

This doesn't fix anything afaict, but will be needed for future
changes.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 465de47bad70 ("anv: associate vulkan formats with aspects")
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoradv: put back VGT_FLUSH at ring init on gfx10
Dave Airlie [Thu, 18 Jul 2019 06:12:53 +0000 (16:12 +1000)]
radv: put back VGT_FLUSH at ring init on gfx10

I can find no evidence that removing this is a good idea.

Fixes: 9b116173b6a ("radv: do not emit VGT_FLUSH on GFX10")
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agosoftpipe: Clamp border colors when needed
Gert Wollny [Tue, 16 Jul 2019 06:05:20 +0000 (08:05 +0200)]
softpipe: Clamp border colors when needed

unorm and snorm require that the border color values are clamped, so when
picking the sampler view copy/clamp the border color from the sampler and
use these adjusted values.

Fixes:

  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_compressed_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_snorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_srgb_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_unorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_compressed_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_snorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_srgb_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_unorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_unorm_depth
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_unorm_depth_uint_stencil_sample_depth

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
5 years agosoftpipe: set a lower minimum clamp value for texture coordinate border clamp
Gert Wollny [Mon, 15 Jul 2019 10:51:06 +0000 (12:51 +0200)]
softpipe: set a lower minimum clamp value for texture coordinate border clamp

The value of -0.5f is not small enough to produce negative coordinates,
so lower the minimum clamp value to -1.0f. This fixes a number of tests
from
   dEQP-GLES31.functional.texture.border_clamp.*

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
5 years agosoftpipe: Correct repeat-mirror evaluation
Gert Wollny [Mon, 15 Jul 2019 11:46:53 +0000 (13:46 +0200)]
softpipe: Correct repeat-mirror evaluation

when mirroring the texture corrdinates the indices must be mirrored as
well and the half pixel shift must be applied in reverse.

Fixes a number of tests from:
  dEQP-GLES31.functional.texture.gather.offset.*
  dEQP-GLES31.functional.texture.gather.offsets.*

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
5 years agosoftpipe: Also mark textures as dirty when updating the framebuffer state
Gert Wollny [Fri, 24 May 2019 10:08:26 +0000 (12:08 +0200)]
softpipe: Also mark textures as dirty when updating the framebuffer state

At this point all the draw caches are flushed to the old attached textures,
so the read caches of these textures will need to be updated too.

Fixes:
   dEQP-GLES3.functional.fbo.color.repeated_clear.sample.tex2d.*

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
5 years agoetnaviv: set DITHER_MODE
Jonathan Marek [Fri, 12 Jul 2019 23:26:32 +0000 (19:26 -0400)]
etnaviv: set DITHER_MODE

This fixes a rendering glitch observed in SDL testscale test, where alpha
blending samples with value (1.0, 1.0, 1.0, 0.0) whitens the target instead
of having no effect.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
5 years agoetnaviv: update headers from rnndb
Jonathan Marek [Fri, 12 Jul 2019 23:25:33 +0000 (19:25 -0400)]
etnaviv: update headers from rnndb

Update to etna_viv commit a16a418.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
5 years agoetnaviv: fix blend color on newer GPUs
Jonathan Marek [Wed, 3 Jul 2019 18:06:17 +0000 (14:06 -0400)]
etnaviv: fix blend color on newer GPUs

Newer GPUs use the half float ALPHA_COLOR_EXT register.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
5 years agoetnaviv: fix alpha blending cases
Jonathan Marek [Wed, 3 Jul 2019 18:04:20 +0000 (14:04 -0400)]
etnaviv: fix alpha blending cases

We need to check rgb_func/alpha_func when determining if blend or separate
alpha is required.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
5 years agoetnaviv: fix polygon offset
Jonathan Marek [Wed, 3 Jul 2019 18:01:33 +0000 (14:01 -0400)]
etnaviv: fix polygon offset

Dividing the fui result by 65535 is obviously wrong, and from testing, on
GC7000L at least there is no division by 65535.

Fixes dEQP-GLES2.functional.polygon_offset.fixed16_displacement_with_units

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
5 years agoradv: dont store disasm string unless keep_shader_info flag set
Timothy Arceri [Wed, 17 Jul 2019 04:20:55 +0000 (14:20 +1000)]
radv: dont store disasm string unless keep_shader_info flag set

This fixes the memory use regression from bug 111107.

Fixes: 726a31df705 ("radv: Add the concept of radv shader binaries.")
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111107

5 years agoradv/gfx10: set the pgm rsrc3/4 regs using index sh reg set
Dave Airlie [Tue, 16 Jul 2019 05:21:56 +0000 (15:21 +1000)]
radv/gfx10: set the pgm rsrc3/4 regs using index sh reg set

This is ported from AMDVLK, it's probably not requires unless
we want to use "real time queues", but it might be nice to just have
in place.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agoradv: use correct register setter for ngg hw addr
Dave Airlie [Wed, 17 Jul 2019 04:55:52 +0000 (14:55 +1000)]
radv: use correct register setter for ngg hw addr

this shouldn't matter, but it's good to be correct.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agofreedreno/a6xx: Drop the WFI in the program update stateobj.
Eric Anholt [Wed, 17 Jul 2019 19:56:12 +0000 (12:56 -0700)]
freedreno/a6xx: Drop the WFI in the program update stateobj.

Rob Clark thinks this was likely a workaround for our const buffer
update bugs, and now that it's passing tests, we should be able to
drop it.

renderdoc-traces results:

traces/android/clashofclans.rdc:  +6.1% +/-   1.1%
traces/android/candycrush.rdc:    +5.2% +/-   1.6%

Reviewed-by: Rob Clark <robdclark@gmail.com>
5 years agofreedreno/a6xx: Drop the WFI in constant uploads.
Eric Anholt [Wed, 19 Jun 2019 18:36:29 +0000 (11:36 -0700)]
freedreno/a6xx: Drop the WFI in constant uploads.

Now that the bin vs render constlen is fixed, we can skip these waits.

Improves webgl aquarium performance at 10k fish from 27fps to 33.
Some highlights from renderdoc-traces:

traces/android/minecraft.rdc:             +17.1% +/-   3.4%
traces/glmark2/ideas-speed=duration.rdc:  +11.6% +/-   2.4%
traces/android/candycrush.rdc:            +5.4%  +/-   1.1%
traces/android/clashofclans.rdc:          +4.4%  +/-   1.3%

Reviewed-by: Rob Clark <robdclark@gmail.com>
5 years agofreedreno: Assert that we don't exceed constlen.
Eric Anholt [Wed, 26 Jun 2019 22:35:44 +0000 (15:35 -0700)]
freedreno: Assert that we don't exceed constlen.

We actually could go up to vs->constlen in the binning shader on a6xx,
but for sanity let's make sure that we're always under constlen.  This
would have caught the bug fixed in 572c76fd8826 ("freedreno: Clamp UBO
uploads to the constlen decided by the shader.")

Reviewed-by: Rob Clark <robdclark@gmail.com>
5 years agofreedreno: Fix more constlen overflows.
Eric Anholt [Wed, 26 Jun 2019 23:47:10 +0000 (16:47 -0700)]
freedreno: Fix more constlen overflows.

Fixes constlen overflow in
dEQP-GLES31.functional.shaders.builtin_var.compute.num_work_groups and
dEQP-GLES31.functional.image_load_store.buffer.image_size.readonly_32
and probably others.

Reviewed-by: Rob Clark <robdclark@gmail.com>
5 years agofreedreno: Drop stale comment about skipping uploads.
Eric Anholt [Wed, 19 Jun 2019 19:00:57 +0000 (12:00 -0700)]
freedreno: Drop stale comment about skipping uploads.

We already skip the upload if it's unused, due to the constlen >
offset check.

Reviewed-by: Rob Clark <robdclark@gmail.com>
5 years agovirgl: Set meta data for textures from handle.
Lepton Wu [Wed, 17 Jul 2019 17:02:20 +0000 (10:02 -0700)]
virgl: Set meta data for textures from handle.

The set of meta data was removed by commit 8083464. It broke lots of
dEQP tests when running with pbuffer surface type.

Fixes: 80834640137 ("virgl: remove dead code")
Signed-off-by: Lepton Wu <lepton@chromium.org>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
5 years agoradv: Only save the descriptor set if we have one.
Bas Nieuwenhuizen [Wed, 17 Jul 2019 00:58:59 +0000 (02:58 +0200)]
radv: Only save the descriptor set if we have one.

After reset, if valid does not contain the relevant bit the descriptor
can be != NULL but still not be valid.

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
5 years agoanv: report timestampComputeAndGraphics true
Lionel Landwerlin [Wed, 17 Jul 2019 05:46:53 +0000 (08:46 +0300)]
anv: report timestampComputeAndGraphics true

Spec says :

   "timestampComputeAndGraphics specifies support for timestamps on all
    graphics and compute queues. If this limit is set to VK_TRUE, all
    queues that advertise the VK_QUEUE_GRAPHICS_BIT or
    VK_QUEUE_COMPUTE_BIT in the VkQueueFamilyProperties::queueFlags
    support VkQueueFamilyProperties::timestampValidBits of at least 36."

On gen7+ this should be true (we only have 32bits of timestamp on
gen6 and below).

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 802f00219addb3 ("anv/device: Update features and limits")
Reported-by: Timothy Strelchun <timothy.strelchun@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoiris: Enable fast clears on other miplevels and layers than 0.
Rafael Antognolli [Tue, 25 Jun 2019 21:14:22 +0000 (14:14 -0700)]
iris: Enable fast clears on other miplevels and layers than 0.

Until now we only supported fast clear colors on the first miplevel and
layer. The main reason for it is that we can't have different fast clear
values at different levels/layers, since the surface state only supports
one clear value.

We can, however, enable it if we make sure we only use the same value
for all levels/layers, and if one of them changes, we resolve all the
others. We already do that for depth fast clears so hopefully it will be
fine for color fast clears too.

v2: Add check for partial clear too (Ken).

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
5 years agoiris: Allow resolving clear color of CCS_D surfaces.
Rafael Antognolli [Wed, 3 Jul 2019 16:37:47 +0000 (09:37 -0700)]
iris: Allow resolving clear color of CCS_D surfaces.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoiris: Make iris_has_color_unresolved non-static
Kenneth Graunke [Wed, 17 Jul 2019 20:43:04 +0000 (13:43 -0700)]
iris: Make iris_has_color_unresolved non-static

We want to use this in the transfer code and possibly for fast clears.

5 years agobroadcom: Move v3d_get_device_info to common
Andreas Bergmeier [Sun, 14 Jul 2019 20:23:15 +0000 (22:23 +0200)]
broadcom: Move v3d_get_device_info to common

In common we can use implementation for Vulkan.

5 years agonir/large_constants: Use dominance information to find more constants
Caio Marcelo de Oliveira Filho [Wed, 5 Jun 2019 20:08:56 +0000 (13:08 -0700)]
nir/large_constants: Use dominance information to find more constants

Relax the restriction that all the writes need to be in the first
block: now accept variables that have all the writes in the same
block, and all the reads are dominated by that block.

This let the pass identify large constants that are local to a helper
function.  The writes will be at the place that the function is
inlined, possibly not in the first block (but still all in the same
block).

Results for vkpipeline-db in SKL:

total instructions in shared programs: 3624891 -> 3623145 (-0.05%)
instructions in affected programs: 79416 -> 77670 (-2.20%)
helped: 16
HURT: 0

total cycles in shared programs: 1458149667 -> 1458147273 (<.01%)
cycles in affected programs: 30154164 -> 30151770 (<.01%)
helped: 14
HURT: 2

total loops in shared programs: 2437 -> 2437 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0

total spills in shared programs: 8813 -> 8745 (-0.77%)
spills in affected programs: 2894 -> 2826 (-2.35%)
helped: 8
HURT: 0

total fills in shared programs: 23470 -> 23392 (-0.33%)
fills in affected programs: 12248 -> 12170 (-0.64%)
helped: 6
HURT: 2

LOST:   0
GAINED: 0

Results for shader-db in SKL with Iris:

total instructions in shared programs: 15379442 -> 15379392 (<.01%)
instructions in affected programs: 837 -> 787 (-5.97%)
helped: 2
HURT: 2
helped stats (abs) min: 27 max: 27 x̄: 27.00 x̃: 27
helped stats (rel) min: 10.47% max: 10.67% x̄: 10.57% x̃: 10.57%
HURT stats (abs)   min: 2 max: 2 x̄: 2.00 x̃: 2
HURT stats (rel)   min: 1.23% max: 1.23% x̄: 1.23% x̃: 1.23%
95% mean confidence interval for instructions value: -39.14 14.14
95% mean confidence interval for instructions %-change: -15.51% 6.17%
Inconclusive result (value mean confidence interval includes 0).

total loops in shared programs: 4880 -> 4880 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0

total cycles in shared programs: 370677237 -> 370676567 (<.01%)
cycles in affected programs: 17852 -> 17182 (-3.75%)
helped: 2
HURT: 1
helped stats (abs) min: 338 max: 356 x̄: 347.00 x̃: 347
helped stats (rel) min: 13.98% max: 14.64% x̄: 14.31% x̃: 14.31%
HURT stats (abs)   min: 24 max: 24 x̄: 24.00 x̃: 24
HURT stats (rel)   min: 0.18% max: 0.18% x̄: 0.18% x̃: 0.18%

total spills in shared programs: 11772 -> 11772 (0.00%)
spills in affected programs: 0 -> 0
helped: 0
HURT: 0

total fills in shared programs: 24948 -> 24948 (0.00%)
fills in affected programs: 0 -> 0
helped: 0
HURT: 0

LOST:   0
GAINED: 0

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agointel/fs: Use a strided MOV instead of a conversion for load_* destinations
Jason Ekstrand [Sat, 13 Jul 2019 23:35:20 +0000 (18:35 -0500)]
intel/fs: Use a strided MOV instead of a conversion for load_* destinations

In many cases, the compiler can just copy-prop the strided MOV whereas
the conversion is a bit trickier.  This cuts 5% of the instructions off
of one particular Vulkan CTS test which does lots of load_ssbo.

Reviewed-by: Matt Turner <mattst88@gmail.com>
5 years agonir/algebraic: Optimize comparisons and up-casts
Jason Ekstrand [Sat, 13 Jul 2019 04:26:59 +0000 (23:26 -0500)]
nir/algebraic: Optimize comparisons and up-casts

These seem like obvious enough optimizations in the world of multiple
integer bit sizes.  The only known thing which hits these at the moment
is some Vulkan CTS tests for 16-bit SSBO values which like to up-cast
and check for equality.  However, it's something that's bound to come up
as we start seeing more integers in shaders.

The optimizations of comparisons of casted values with constants are
something which we would ideally do with range analysis.  However,
lacking that, we can do it in opt_algebraic as long as one side is a
constant.

In dEQP-VK.ssbo.phys.layout.random.16bit.scalar.13, this commit, along
with the previous commit, reduce the number of instructions emitted on
Skylake from 55328 to 44546, a reduction of 20%.

Acked-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
5 years agonir/algebraic: Optimize comparing unpacked values
Jason Ekstrand [Sat, 13 Jul 2019 04:26:48 +0000 (23:26 -0500)]
nir/algebraic: Optimize comparing unpacked values

We could, in theory, add the same optimization for 64-bit unpack
operations but that's likely to fight with 64-bit integer lowering on
platforms which require it so it will require more infrastructure before
that will be a good idea.

Reviewed-by: Matt Turner <mattst88@gmail.com>
5 years agonir/algebraic: Print out the list of transforms in the C file
Jason Ekstrand [Sat, 13 Jul 2019 15:57:24 +0000 (10:57 -0500)]
nir/algebraic: Print out the list of transforms in the C file

This helps greatly when debugging algebraic transform generators because
you can now actually see the output and verify that your transforms are
getting generated.

Acked-by: Matt Turner <mattst88@gmail.com>
5 years agointel/fs: Properly stride NULL replacement regs in DCE
Jason Ekstrand [Sat, 13 Jul 2019 17:06:00 +0000 (12:06 -0500)]
intel/fs: Properly stride NULL replacement regs in DCE

This fixes some validation errors generated by certain D->W conversions
but is likely not a full solution.  Calculating an actual register
stride is a far more complex problem in general and should probably be
handled by the brw_fs_generator.

Reviewed-by: Matt Turner <mattst88@gmail.com>
5 years agonir: Fix nir_lower_alu_to_scalar's instr filtering.
Eric Anholt [Tue, 16 Jul 2019 19:38:10 +0000 (12:38 -0700)]
nir: Fix nir_lower_alu_to_scalar's instr filtering.

It was checking if the dest or src[0] SSA values were vectors, rather than
whether the ALU op was using the source as a vector resulting in a
nir_fdot4 making it through to vc4 and v3d:

vec1 32 ssa_6 = fdot4 ssa_4.xxxx, ssa_5

Fixes: c1cffa4249ca ("nir/alu_to_scalar: Use the new NIR lowering framework")
v2: Use Jason's recommendation to look at input_sizes.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agopanfrost: Merge varyings_mem into transient buffers
Alyssa Rosenzweig [Tue, 16 Jul 2019 18:36:11 +0000 (11:36 -0700)]
panfrost: Merge varyings_mem into transient buffers

Theoretically we would like these split since varyings can have
specially optimized flags (no map, coherent local). For now, since
neither of these flags is particularly meaningful right now, merge them
together instead of special casing varyings_mem.

Saves upwards of 64MB of RAM per context.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agovulkan/wsi: update swapchain status on vkQueuePresent
Lionel Landwerlin [Mon, 15 Jul 2019 06:01:20 +0000 (09:01 +0300)]
vulkan/wsi: update swapchain status on vkQueuePresent

With the following chain of events :

   vkQueuePresent()
   <- Surface resize
   vkQueuePresent()

We should be able to report SUBOPTIMAL or OUT_OF_DATE on the second
vkQueuePresent() call. Currently we only look at X11 events in the
vkAcquireNextImage() path so we're not able to report this.

This change checks the queue of events and process any available ones
to update the swapchain status.

v2: Be consistent about reporting the current error state of the
    swapchain (Jason)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111097
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoradv: add an option for disabling NGG on GFX10
Samuel Pitoiset [Wed, 17 Jul 2019 12:19:10 +0000 (14:19 +0200)]
radv: add an option for disabling NGG on GFX10

Will be useful for testing the legacy path.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agosoftpipe: pass stream-out targets to draw-module early
Erik Faye-Lund [Tue, 7 May 2019 10:13:43 +0000 (12:13 +0200)]
softpipe: pass stream-out targets to draw-module early

This is essensially a port of ed53e61bec9 from LLVMpipe to softpipe,
as it makes things a bit simpler and more performant.

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-By: Gert Wollny <gert.wollny@collabora.com>
5 years agospirv_extensions: i965: initialize SPIR-V extensions
Alejandro Piñeiro [Sat, 14 Oct 2017 08:52:51 +0000 (10:52 +0200)]
spirv_extensions: i965: initialize SPIR-V extensions

v2: Rebase update after changes on previous patches.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
5 years agospirv_extensions: add spirv_supported_extensions on gl_constants
Alejandro Piñeiro [Fri, 13 Oct 2017 18:43:01 +0000 (20:43 +0200)]
spirv_extensions: add spirv_supported_extensions on gl_constants

We can use it to get real values for ARB_spirv_extensions methods.

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Arcady Goldmints-Orlov <agoldmints@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
5 years agospirv_extensions: define spirv_extensions_supported
Alejandro Piñeiro [Wed, 19 Jun 2019 18:58:30 +0000 (13:58 -0500)]
spirv_extensions: define spirv_extensions_supported

Add a struct to maintain which SPIR-V extensions are supported, and an
utility method to initialize it based on
nir_spirv_supported_capabilities.

v2:
  * Fixing code style (Ian Romanick)
  * Adding a prefix (spirv) to fill_supported_spirv_extensions (Ian Romanick)

v3: rebase update (nir_spirv_supported_extensions renamed)

v4: include AMD_gcn_shader support

v5: move spirv_fill_supported_spirv_extensions to
    src/mesa/main/spirv_extensions.c

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Arcady Goldmints-Orlov <agoldmints@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
5 years agospirv_extensions: add list of extensions and to_string method
Alejandro Piñeiro [Fri, 13 Oct 2017 14:17:51 +0000 (16:17 +0200)]
spirv_extensions: add list of extensions and to_string method

Ideally this should be generated somehow. One option would be gather
all the extension dependencies listed on the core grammar, but there
would be the possibility of not including some of the extensions.

Note that spirv-tools is doing it just slightly better, as it has a
hardcoded list of extensions manually took from the registry, that
they parse to get the enum and the to_string method (see
generate_grammar_tables.py).

v2:
  * Use a macro to improve readability. (Tapani Pälli)
  * Add unreachable on the switch, no default (Eric Engestrom)
  * No typedef enum (Ian Romanick)
  * Sort extensions names (Ian Romanick)
  * Don't add extensions unlikely to be supported by Mesa at any point
    (Ian Romanick)

v3: rebase update

v4: Include AMD_gcn_shader

v5: move spirv_extensions_to_string to src/mesa/main/spirv_extensions.c

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Arcady Goldmints-Orlov <agoldmints@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
5 years agospirv_extensions: add GL_ARB_spirv_extensions boilerplate
Alejandro Piñeiro [Tue, 10 Oct 2017 14:09:25 +0000 (16:09 +0200)]
spirv_extensions: add GL_ARB_spirv_extensions boilerplate

v2:
  * Mention extension gap at gl_API.xml (Emil Velikov)
  * Bail with INVALID_ENUM if extension not available on getStringi (Emil Velikov)
  * Use EXTRA_EXT macro when defining the extension at
    get.c/get_hash_params.py (Emil Velikov)
  * Rename source files (spirvextensions.[ch] -> spirv_extensions.[ch]) (Ian)

v3:
  * Fix GL_PROGRAM_BINARY_FORMATS glGet query, broken by error on a
    previous rebase

v4:
   * Fix rebase conflicts on getstring.c after
     GL_SHADING_LANGUAGE_VERSION query was added

v5:
   * Remove src/mapi/glapi/gen/Makefile.am as it no longer exists in
     master

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Arcady Goldmints-Orlov <agoldmints@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
5 years agoradv/gfx10: implement VK_EXT_post_depth_coverage
Samuel Pitoiset [Tue, 16 Jul 2019 15:11:50 +0000 (17:11 +0200)]
radv/gfx10: implement VK_EXT_post_depth_coverage

I did implement this extension a while ago but it didn't work
on pre GFX10 for some reasons. Now all CTS pass.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agoradv/gfx10: disable the TC compat zrange workaround
Samuel Pitoiset [Tue, 16 Jul 2019 15:35:00 +0000 (17:35 +0200)]
radv/gfx10: disable the TC compat zrange workaround

Unnecessary.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agoradv/gfx10: fallback to the legacy path if tess and extreme geometry
Samuel Pitoiset [Tue, 16 Jul 2019 14:39:16 +0000 (16:39 +0200)]
radv/gfx10: fallback to the legacy path if tess and extreme geometry

This is unsupported and hangs.

This fixes GPU hangs with
dEQP-VK.tessellation.geometry_interaction.limits.output_required_*.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agoradv/gfx10: always build the GS copy shader but uses it on-demand
Samuel Pitoiset [Tue, 16 Jul 2019 14:39:15 +0000 (16:39 +0200)]
radv/gfx10: always build the GS copy shader but uses it on-demand

It should be possible to build it on-demand too but it requires
more work. On GFX10, the GS copy shader is required when tess
is enabled with extreme geometry.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agosoftpipe: Remove unused static function
Gert Wollny [Thu, 2 May 2019 13:54:55 +0000 (15:54 +0200)]
softpipe: Remove unused static function

Thanks to Eric Engestrom for pointing out that there was something wrong
with that function.

Fixes: 724a73509e1bc1ce3abf9500e457bb2911b642db
  softpipe: Prepare handling explicit gradients

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agospirv: Bail when we see CounterBuffer decoration
Caio Marcelo de Oliveira Filho [Wed, 10 Jul 2019 16:00:47 +0000 (09:00 -0700)]
spirv: Bail when we see CounterBuffer decoration

This decoration can be ignored, so we can just skip the next steps.
Otherwise we'd have to also handle it in apply_var_decoration.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoiris: Drop copy and pasted iris_timebase_scale
Kenneth Graunke [Mon, 15 Jul 2019 18:51:14 +0000 (11:51 -0700)]
iris: Drop copy and pasted iris_timebase_scale

Lionel moved brw_timebase_scale to gen_device_info_timebase_scale a few
months ago, so we should just use that, and not our own copy in iris.

5 years agonir/regs_to_ssa: Handle regs in phi sources properly
Jason Ekstrand [Fri, 12 Jul 2019 16:01:40 +0000 (11:01 -0500)]
nir/regs_to_ssa: Handle regs in phi sources properly

Sources of phi instructions act as if they occur at the very end of the
predecessor block not the block in which the phi lives.  In order to
handle them correctly, we have to skip phi sources on the normal
instruction walk and handle them as a separate walk over the successor
phis.  While registers in phi instructions is a bit of an oddity it can
happen when we temporarily go out-of-SSA for control-flow manipulations.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111075
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
5 years agospirv: Add a warning for ArrayStride on arrays of blocks
Jason Ekstrand [Wed, 10 Jul 2019 22:55:44 +0000 (17:55 -0500)]
spirv: Add a warning for ArrayStride on arrays of blocks

It's disallowed according to the SPIR-V spec or at least I think that's
what the spec says.  It's in a section explicitly about explicit layout
of things in the StorageBuffer, Uniform, and PushConstant storage
classes so it's not 100% clear that it applies with other storage
classes.  However, it seems like it should apply in general and
violating it can trigger (fairly harmless) asserts in NIR.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
5 years agoanv: Increase state allocation size limit to 2MB
Caio Marcelo de Oliveira Filho [Fri, 12 Jul 2019 18:33:38 +0000 (11:33 -0700)]
anv: Increase state allocation size limit to 2MB

When running on ICL the
dEQP-VK.ssbo.phys.layout.random.16bit.scalar.13 needs more than 1M for
the shader, so bump it.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agometa: leaking of BO with DrawPixels
Yevhenii Kolesnikov [Tue, 9 Jul 2019 12:11:15 +0000 (15:11 +0300)]
meta: leaking of BO with DrawPixels

ctx->Unpack.BufferObj wasn't unreferenced.

Fixes: d492e7b0171 (meta: Fix invalid PBO access from DrawPixels when
trying to just alloc.)
CC: Eric Anholt <eric@anholt.net>
Signed-off-by: Yevhenii Kolesnikov <yevhenii.kolesnikov@globallogic.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agoswrast: Move _mesa_format_pack_colormask() to the only caller.
Eric Anholt [Tue, 2 Jul 2019 00:20:55 +0000 (17:20 -0700)]
swrast: Move _mesa_format_pack_colormask() to the only caller.

This avoids needing format_pack to have access to the GLenum return
functions for mesa_format.  It seems like an odd function and unlikely
to be reused.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Give _mesa_format_get_color_encoding a clearer name.
Eric Anholt [Mon, 1 Jul 2019 22:43:19 +0000 (15:43 -0700)]
mesa: Give _mesa_format_get_color_encoding a clearer name.

It only returned one of two values.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Drop redundant checks for sRGB before sRGB to linear conversion.
Eric Anholt [Mon, 1 Jul 2019 22:38:56 +0000 (15:38 -0700)]
mesa: Drop redundant checks for sRGB before sRGB to linear conversion.

_mesa_get_srgb_format_linear() just returns the original format if it
wasn't sRGB.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Fold _mesa_unpack_depth_stencil_row() into its only caller.
Eric Anholt [Mon, 1 Jul 2019 23:45:32 +0000 (16:45 -0700)]
mesa: Fold _mesa_unpack_depth_stencil_row() into its only caller.

This was the last bit of gl.h usage in format packing.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Convert format_pack/unpack off of GL types.
Eric Anholt [Mon, 1 Jul 2019 23:27:24 +0000 (16:27 -0700)]
mesa: Convert format_pack/unpack off of GL types.

Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Port format_pack/unpack off of _mesa_problem().
Eric Anholt [Mon, 1 Jul 2019 23:34:23 +0000 (16:34 -0700)]
mesa: Port format_pack/unpack off of _mesa_problem().

unreachable() should be plenty of debug for these.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Mostly switch Mesa format info off of GL types other than GLenum.
Eric Anholt [Mon, 1 Jul 2019 22:23:19 +0000 (15:23 -0700)]
mesa: Mostly switch Mesa format info off of GL types other than GLenum.

I'm considering moving most of this code to src/util/, and I want that
code to not expose GL types in its interfaces.

Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Rename gl_pack typedefs to mesa_pack.
Eric Anholt [Mon, 1 Jul 2019 21:29:45 +0000 (14:29 -0700)]
mesa: Rename gl_pack typedefs to mesa_pack.

These are packing mesa formats, not a GL format/type.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agomesa: Rename gl_format_info to mesa_format_info.
Eric Anholt [Mon, 1 Jul 2019 21:07:30 +0000 (14:07 -0700)]
mesa: Rename gl_format_info to mesa_format_info.

It's about MESA_FORMATs, after all.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agostate_tracker: Move the format test out to be an actual unit test.
Eric Anholt [Mon, 1 Jul 2019 20:06:09 +0000 (13:06 -0700)]
state_tracker: Move the format test out to be an actual unit test.

We want errors in the table to show up as unit test failures in MRs.
Also keeps unit test code out of the built drivers.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agou_format: Remove pointless comments.
Eric Anholt [Mon, 1 Jul 2019 19:38:41 +0000 (12:38 -0700)]
u_format: Remove pointless comments.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agosrc/util: Switch _mesa_half_to_float() to u_half.h's version.
Eric Anholt [Thu, 27 Jun 2019 22:53:53 +0000 (15:53 -0700)]
src/util: Switch _mesa_half_to_float() to u_half.h's version.

The two implementations differ across the entire input range only in
that u_half.h preserves mantissa bits for NaNs.  The u_half.h version
shaves 15% off of the text size of half_float.o.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agou_half_test: Turn it into an actual unit test.
Eric Anholt [Thu, 27 Jun 2019 22:46:28 +0000 (15:46 -0700)]
u_half_test: Turn it into an actual unit test.

You could break the test and meson test wouldn't complain, since we
returned success either way.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
5 years agoandroid: radv/gfx10: generate gfx10_format_table.h
Mauro Rossi [Sun, 7 Jul 2019 23:17:22 +0000 (01:17 +0200)]
android: radv/gfx10: generate gfx10_format_table.h

This patch adds the missing building rules for Android,
to avoid following building errors:

In file included from external/mesa/src/amd/vulkan/radv_debug.c:35:
In file included from external/mesa/src/amd/vulkan/radv_debug.h:27:
external/mesa/src/amd/vulkan/radv_private.h:95:10:
fatal error: 'gfx10_format_table.h' file not found
         ^~~~~~~~~~~~~~~~~~~~~~
1 error generated.

In file included from external/mesa/src/amd/vulkan/radv_android.c:31:
external/mesa/src/amd/vulkan/radv_private.h:95:10:
fatal error: 'gfx10_format_table.h' file not found
         ^~~~~~~~~~~~~~~~~~~~~~
1 error generated.

Fixes: 3dc5ec5d16 ("radv/gfx10: generate gfx10_format_table.h")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
5 years agomesa/st: add sampler uniforms
Rob Clark [Mon, 15 Jul 2019 16:05:55 +0000 (09:05 -0700)]
mesa/st: add sampler uniforms

Add sampler uniforms for the UV plane(s), so driver can count the
uniforms and get the correct sampler count.

Fixes lowered YUV on a6xx which actually wants to know # of samplers.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agoegl/android: handle multi-fd native windows
Rob Clark [Fri, 12 Jul 2019 16:17:30 +0000 (09:17 -0700)]
egl/android: handle multi-fd native windows

We can hit multi-fd EGL_NATIVE_BUFFER_ANDROID case when the native
android buffer is YUV.  So we need to handle that.

Currently this went unnoticed because, even though we have two or
three fd's for YUV native android buffers, they all reference the
same backing buffer.  But we really shouldn't rely on that.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agost,i965: Stop looping on 64-bit lowering
Jason Ekstrand [Thu, 11 Jul 2019 21:59:31 +0000 (16:59 -0500)]
st,i965: Stop looping on 64-bit lowering

Now that the 64-bit lowering passes do a complete lowering in one go, we
don't need to loop anymore.  We do, however, have to ensure that int64
lowering happens after double lowering because double lowering can
produce int64 ops.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_doubles: Handle fdiv and fsub directly
Jason Ekstrand [Thu, 11 Jul 2019 22:06:31 +0000 (17:06 -0500)]
nir/lower_doubles: Handle fdiv and fsub directly

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_doubles: Use the new NIR lowering framework
Jason Ekstrand [Thu, 11 Jul 2019 21:54:01 +0000 (16:54 -0500)]
nir/lower_doubles: Use the new NIR lowering framework

One advantage of this is that we no longer need to run in a loop because
the new framework handles lowering instructions added by lowering.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_doubles: Use "alu" for the nir_alu_instr
Jason Ekstrand [Thu, 11 Jul 2019 21:46:19 +0000 (16:46 -0500)]
nir/lower_doubles: Use "alu" for the nir_alu_instr

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_int64: Use the core NIR lowering framework
Jason Ekstrand [Thu, 11 Jul 2019 21:31:49 +0000 (16:31 -0500)]
nir/lower_int64: Use the core NIR lowering framework

One advantage of this is that we no longer need to run in a loop because
the new framework handles lowering instructions added by lowering.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/alu_to_scalar: Use the new NIR lowering framework
Jason Ekstrand [Thu, 11 Jul 2019 20:20:39 +0000 (15:20 -0500)]
nir/alu_to_scalar: Use the new NIR lowering framework

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/alu_to_scalar: Use "alu" as the name for the nir_alu_instr
Jason Ekstrand [Thu, 11 Jul 2019 20:19:17 +0000 (15:19 -0500)]
nir/alu_to_scalar: Use "alu" as the name for the nir_alu_instr

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_system_values: Support lowering more intrinsics
Jason Ekstrand [Fri, 12 Jul 2019 20:26:03 +0000 (15:26 -0500)]
nir/lower_system_values: Support lowering more intrinsics

Instead of only lowering system from variables, lower most to intrinsics
and let the lowering framework immediately lower the intrinsic.  This
will result in a bit more instruction churn but it means that NIR code
builders can just use intrinsics instead of everything having to go
through variables.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_system_values: Drop the context-aware builder functions
Jason Ekstrand [Thu, 11 Jul 2019 19:01:20 +0000 (14:01 -0500)]
nir/lower_system_values: Drop the context-aware builder functions

Instead of having context-aware builder functions, just provide lowering
for the system value intrinsics and let nir_shader_lower_instructions
handle the recursion for us.  This makes everything a bit simpler and
means that the lowering can also be used if something comes in as a
system value intrinsic rather than a load_deref.

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_system_values: Use the new generic NIR lowering helpers
Jason Ekstrand [Thu, 11 Jul 2019 18:30:03 +0000 (13:30 -0500)]
nir/lower_system_values: Use the new generic NIR lowering helpers

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir/lower_subgroups: Use the new generic NIR lowering helpers
Jason Ekstrand [Thu, 11 Jul 2019 18:04:05 +0000 (13:04 -0500)]
nir/lower_subgroups: Use the new generic NIR lowering helpers

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir: Add some generic helpers for writing lowering passes
Jason Ekstrand [Thu, 11 Jul 2019 18:00:42 +0000 (13:00 -0500)]
nir: Add some generic helpers for writing lowering passes

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agonir: Add a helper for fetching the SSA def from an instruction
Jason Ekstrand [Thu, 11 Jul 2019 20:05:27 +0000 (15:05 -0500)]
nir: Add a helper for fetching the SSA def from an instruction

Reviewed-by: Eric Anholt <eric@anholt.net>
5 years agopandecode: Add more addresses to trace
Tomeu Vizoso [Fri, 12 Jul 2019 14:42:52 +0000 (16:42 +0200)]
pandecode: Add more addresses to trace

When debugging, we're given the fault_pointer unresolved, so it is
helpful to have more context in the decode.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
5 years agopanfrost: Use 64-bit descriptors globally
Tomeu Vizoso [Thu, 11 Jul 2019 06:06:41 +0000 (08:06 +0200)]
panfrost: Use 64-bit descriptors globally

Midgard supports two modes of operation, 32-bit mode and 64-bit mode.
The GPU is natively 64-bit, but job descriptors can be submitted in
32-bit mode. Among other changes, 32-bit mode shortens pointer sizes to
use 32-bit pointers rather than the full 64-bit range.

The blob decides which mode to use based on the CPU bitness, so an armhf
system uses 32-bit descriptors and an aarch64 system uses 64-bit
descriptors. For a while, we mimicked this, bu inevitably this caused
the 32-bit support to lag behind as our reference platform is 64-bit.

To combat the code staleness, we traced an older GPU paired with a 64-bit
CPU (the Midgard T720 on-board the sunxi H64). From there, we could tell
which fields were really about hardware and which fields were simply
reflections of the descriptor bitness.

From there, we decided to remove support for 32-bit descriptors
entirely, using 64-bit descriptors unconditionally. There is minimal
performance penalty for this in practice, and it allows us to unify
these disparate code paths. This fixes:

   - T860 + armhf
   - T820 + armhf
   - T760 + aarch64

And will help bringup of 1st/2nd generation Midgard regardless of CPU.

[Work done by Tomeu. Commit message written by Alyssa.]

v2: Add comments preserving information about the old behaviour for
future reference. Fix a compiler warning. (Alyssa)

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agoanv: Account for dynamic stencil write disables in the PMA fix
Jason Ekstrand [Mon, 15 Jul 2019 22:14:26 +0000 (17:14 -0500)]
anv: Account for dynamic stencil write disables in the PMA fix

In 6ce8592836b8 we started looking at the dynamic stencil state and
disabling stencil writes when the stencil mask is zero.  Unfortunately,
we never updated the PMA fix code accordingly so 3DSTATE_WM_DEPTH_STENCIL
and the PMA fix were getting out-of-sync causing hangs.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109203
Fixes: 6ce8592836 "anv: Disable stencil writes when both write..."
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agopanfrost: Implement opportunistic AFBC
Alyssa Rosenzweig [Mon, 15 Jul 2019 21:15:24 +0000 (14:15 -0700)]
panfrost: Implement opportunistic AFBC

Rather than hardcoding a BO layout at creation-time, we implement the
ability to hint layouts at various points in a BO's lifetime,
potentially reallocating and switching layouts if it's heuristically
deemed useful to do so.

In this patch, we add a simple hinting implementation, opportunistically
compressing FBOs.

Support is hidden behind PAN_MESA_DEBUG=afbc as the implementation is
incomplete (software access to AFBC is unimplemented at the moment) and
therefore would regress significantly.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopanfrost/mfbd: Zero out framebuffer_stride
Alyssa Rosenzweig [Mon, 15 Jul 2019 22:35:30 +0000 (15:35 -0700)]
panfrost/mfbd: Zero out framebuffer_stride

We don't know what this is, so let's not pretend we do.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopanfrost: AFBC buffers must be cache-line aligned
Alyssa Rosenzweig [Mon, 15 Jul 2019 22:34:50 +0000 (15:34 -0700)]
panfrost: AFBC buffers must be cache-line aligned

Fixes a DATA_INVALID_FAULT when AFBC is paried with mipmapping.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopanfrost: Add Z/S and MRT BOs to the job
Alyssa Rosenzweig [Mon, 15 Jul 2019 22:16:08 +0000 (15:16 -0700)]
panfrost: Add Z/S and MRT BOs to the job

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>