platform/upstream/mesa.git
16 months agonir: add nir_[fui]gt_imm and nir_[fui]le_imm helpers
Erik Faye-Lund [Mon, 5 Jun 2023 08:19:58 +0000 (10:19 +0200)]
nir: add nir_[fui]gt_imm and nir_[fui]le_imm helpers

These are similar to the nir_{cmp}_imm variants we already have, except
they negate the condition (apart from equality) and flip the arguments.
The reason we need this, is that we don't have all comparison directions
that would be required to always pass the immediate in the second
argument.

This allows us to create any comparison with an immediate without
having to manually create the immediate value.

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23460>

16 months agoac: fix PIPE_FORMAT_R11G11B10_FLOAT DST_SEL_W
Rhys Perry [Wed, 14 Jun 2023 11:14:19 +0000 (12:14 +0100)]
ac: fix PIPE_FORMAT_R11G11B10_FLOAT DST_SEL_W

Previously, the W component would be incorrect for attributes using this
format when loaded in RADV's vertex shader prologs.

Fixes dEQP-VK.pipeline.fast_linked_library.vertex_input.*b10g11r11*missing_components*

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Fixes: 6a2ada93b49 ("ac: add ac_vtx_format_info")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23645>

16 months agopvr: fix division by block size in blit
Luigi Santivetti [Mon, 5 Jun 2023 22:34:12 +0000 (23:34 +0100)]
pvr: fix division by block size in blit

dEQP-VK.pipeline.monolithic.image.suballocation.\
 sampling_type.combined.view_type.2d.\
 format.etc2_r8g8b8_unorm_block.count_1.size.13x13

Signed-off-by: Luigi Santivetti <luigi.santivetti@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23591>

16 months agopvr: Remove outdated finishme
Karmjit Mahil [Tue, 13 Jun 2023 10:46:20 +0000 (11:46 +0100)]
pvr: Remove outdated finishme

The color attachment load is being handled just below the finishme.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23619>

16 months agopvr: Dedup a check with pvr_is_render_area_tile_aligned()
Karmjit Mahil [Fri, 10 Feb 2023 16:00:04 +0000 (16:00 +0000)]
pvr: Dedup a check with pvr_is_render_area_tile_aligned()

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22657>

16 months agor600/sfn: Downgrade some error message to warning
Gert Wollny [Thu, 15 Jun 2023 05:55:45 +0000 (07:55 +0200)]
r600/sfn: Downgrade some error message to warning

This doesn't seem to be an error, so just print a warning when
warnings are enabled.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23665>

16 months agor600/sfn: Silence warnings "overloaded-virtual"
Gert Wollny [Thu, 15 Jun 2023 05:54:27 +0000 (07:54 +0200)]
r600/sfn: Silence warnings "overloaded-virtual"

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23665>

16 months agoradv: reset some dynamic states when the fragment shader stage is unbound
Samuel Pitoiset [Mon, 12 Jun 2023 06:43:34 +0000 (08:43 +0200)]
radv: reset some dynamic states when the fragment shader stage is unbound

Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23494>

16 months agoradv: fix re-emitting early_z/late_z when the bound PS changes
Samuel Pitoiset [Mon, 12 Jun 2023 06:37:10 +0000 (08:37 +0200)]
radv: fix re-emitting early_z/late_z when the bound PS changes

This state depends on the fragment shader.

Fixes: d740e283e16 ("radv: implement VK_EXT_attachment_feedback_loop_dynamic_state")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23494>

16 months agoradv: fix re-emitting some dynamic states when the previous FS is NULL
Samuel Pitoiset [Wed, 7 Jun 2023 08:26:38 +0000 (10:26 +0200)]
radv: fix re-emitting some dynamic states when the previous FS is NULL

If the previous FS is NULL, some dynamic states still need to be
re-emitted.

Doesn't fix anything known.

Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23494>

16 months agomesa/arbprog: fix compile errors
Christian Gmeiner [Wed, 14 Jun 2023 10:23:39 +0000 (12:23 +0200)]
mesa/arbprog: fix compile errors

When DEBUG_FP is set I see the following compiler errors:

../../src/gitlab_mesa/src/mesa/program/arbprogparse.c: In function '_mesa_parse_arb_fragment_program':
../../src/gitlab_mesa/src/mesa/program/arbprogparse.c:133:4: error: implicit declaration of function '_mesa_print_program'; did you mean '_mesa_parse_arb_program'? [-Werror=implicit-function-declaration]
  133 |    _mesa_print_program(&program->Base);
      |    ^~~~~~~~~~~~~~~~~~~
      |    _mesa_parse_arb_program
../../src/gitlab_mesa/src/mesa/program/arbprogparse.c:133:32: error: 'struct gl_program' has no member named 'Base'
  133 |    _mesa_print_program(&program->Base);
      |                                ^~
cc1: some warnings being treated as errors

Signed-off-by: Christian Gmeiner <cgmeiner@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23644>

16 months agoac/nir/ngg: fix ngg_gs_clear_primflags crash
Qiang Yu [Mon, 12 Jun 2023 09:50:50 +0000 (17:50 +0800)]
ac/nir/ngg: fix ngg_gs_clear_primflags crash

We get current_clear_primflag_idx_var==NULL when stream 0 output
number is known, so output_compile_time_known==true. But we also
need this variable when stream 1~3 output number is unknown or
vertex number is less than a primitive's needs.

Fixes: 60ac5dda82e1 ("ac: Add NIR lowering for NGG GS.")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23614>

16 months agozink: strip format list when disabling mutable during image creation
Mike Blumenkrantz [Tue, 13 Jun 2023 15:35:04 +0000 (11:35 -0400)]
zink: strip format list when disabling mutable during image creation

drivers shouldn't be getting a format list if it won't be used

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23514>

16 months agozink: drop dt checks for mutable format init
Mike Blumenkrantz [Wed, 14 Jun 2023 14:46:01 +0000 (10:46 -0400)]
zink: drop dt checks for mutable format init

these are no longer applicable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23514>

16 months agozink: add srgb mutable for all resources by default
Mike Blumenkrantz [Wed, 7 Jun 2023 18:37:16 +0000 (14:37 -0400)]
zink: add srgb mutable for all resources by default

this should enable compression on more intermediate fb attachments

it also means that VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT can now be set
on images where ZINK_BIND_MUTABLE is not set, so non-resource APIs need
to check ZINK_BIND_MUTABLE

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23514>

16 months agozink: wrap format mismatch checks for blit/surface
Mike Blumenkrantz [Wed, 14 Jun 2023 12:49:57 +0000 (08:49 -0400)]
zink: wrap format mismatch checks for blit/surface

no functional changes

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23514>

16 months agozink: remove redundant conditional in set_sampler_views
Mike Blumenkrantz [Wed, 7 Jun 2023 21:02:35 +0000 (17:02 -0400)]
zink: remove redundant conditional in set_sampler_views

it's redundant, but it checks a different flag so it consumes cycles

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23514>

16 months agocompiler/types: Use hash table pre-hashed functions for type caching
Caio Oliveira [Fri, 26 May 2023 21:05:28 +0000 (14:05 -0700)]
compiler/types: Use hash table pre-hashed functions for type caching

Calculate the hash outside the critical region, then use that both
for search and insertion.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23280>

16 months agocompiler/types: Tidy up the asserts in get_*_instance functions
Caio Oliveira [Sat, 27 May 2023 00:38:13 +0000 (17:38 -0700)]
compiler/types: Tidy up the asserts in get_*_instance functions

Use the local variable in the assertions, move them out the critical region.
No behavior change.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23279>

16 months agocompiler/types: Be consistent when naming array element/size
Caio Oliveira [Fri, 26 May 2023 20:43:05 +0000 (13:43 -0700)]
compiler/types: Be consistent when naming array element/size

The element type passed is different than the array type and it is not
a "base type" in the glsl_type sense, so pick a name that reflects that.
Also stick to a single name for the array_size.

Just renames, no behavior change.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23279>

16 months agonir_lower_returns: Mark assert-only var as ASSERTED
Jesse Natalie [Tue, 13 Jun 2023 17:04:04 +0000 (10:04 -0700)]
nir_lower_returns: Mark assert-only var as ASSERTED

Fixes: 5d238c0c ("nir_lower_returns: Optimize phis before beginning the pass")
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23634>

16 months agoradv/video: restrict the number of IBs on video related queues.
Dave Airlie [Wed, 14 Jun 2023 02:12:48 +0000 (12:12 +1000)]
radv/video: restrict the number of IBs on video related queues.

The hardware gets given a session context from userspace in each
submission, but if the session context changes the hardware wants
a FENCE to be emitted to know it can give up the current session.

IF a test submits interleaved session ctx access and uses a single
vulkan submit the hardware crashes, unless each IB is submitted
in a separate submission so the fence can be sent.

In theory it could be possible to construct a single command buffer
to trigger this so I do think the hardware should be smarter here.

Should this be fixed in the kernel to always emit a fence between
IBs?

Fixes: dEQP-VK.video.decode.h264_interleaved

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23641>

16 months agorusticl: fix UB in CLProp machinery
LingMan [Wed, 14 Jun 2023 07:02:53 +0000 (09:02 +0200)]
rusticl: fix UB in CLProp machinery

Viewing structs as a collection of u8 is not generally sound. Any padding bytes might be
uninitialized and creating an integer from uninitialized memory constitutes producing an invalid
value, which is instant UB.

Since we only copy these bytes around, the fix is to simply work with MaybeUninit<u8>, which can handle uninitialized memory just fine, instead.

See: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23652>

16 months agorusticl: drop cl_prop_for_type macro
LingMan [Wed, 14 Jun 2023 02:53:19 +0000 (04:53 +0200)]
rusticl: drop cl_prop_for_type macro

There's no reason to differentiate between primitive types and structs here. `cl_prop_for_struct`
can handle primitive types just fine.
Drop `cl_prop_for_type` and rename the existing `cl_prop_for_struct` to `cl_prop_for_type`.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23652>

16 months agorusticl: drop CLProp implementation for String
LingMan [Wed, 14 Jun 2023 06:16:40 +0000 (08:16 +0200)]
rusticl: drop CLProp implementation for String

Route the data to the implementation for &str instead. It works just as fine.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23652>

16 months agorusticl: core: stop using cl_prop from the api module
LingMan [Wed, 14 Jun 2023 05:56:50 +0000 (07:56 +0200)]
rusticl: core: stop using cl_prop from the api module

It's a layering violation and really the wrong tool for the job. Add a new fn to view a given slice
as a &[u8] instead of going though the clprop machinery which creates a new Vec.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23652>

16 months agosvga: fix compute shader type after ntt
Charmaine Lee [Wed, 14 Jun 2023 05:24:32 +0000 (08:24 +0300)]
svga: fix compute shader type after ntt

Reset compute shader type after ntt.

Fixes: 0ac95418048 ("gallium: Drop PIPE_SHADER_CAP_PREFERRED_IR")

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23659>

16 months agorusticl: enforce using unsafe blocks in unsafe functions
Karol Herbst [Wed, 14 Jun 2023 21:06:23 +0000 (23:06 +0200)]
rusticl: enforce using unsafe blocks in unsafe functions

Signed-off-by: Karol Herbst <git@karolherbst.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23660>

16 months agozink: add mem debugging
Mike Blumenkrantz [Tue, 13 Jun 2023 14:58:12 +0000 (10:58 -0400)]
zink: add mem debugging

modeled off turnip's debug infra, this adds debug printing for oom
scenarios

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23653>

16 months agozink: break out vk flag unrolling into util function
Mike Blumenkrantz [Tue, 13 Jun 2023 14:45:38 +0000 (10:45 -0400)]
zink: break out vk flag unrolling into util function

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23653>

16 months agonir/algebraic: Simplify various trivial bfi
Ian Romanick [Tue, 31 Mar 2020 22:37:00 +0000 (15:37 -0700)]
nir/algebraic: Simplify various trivial bfi

These are mostly just obvious patterns that somebody will eventually
want to add.

DG2, Tiger Lake, Ice Lake, Skylake, Broadwell, and Haswell had similar
results (Ice Lake shown)
total instructions in shared programs: 20570033 -> 20570026 (<.01%)
instructions in affected programs: 7363 -> 7356 (-0.10%)
helped: 6 / HURT: 0

total cycles in shared programs: 902118781 -> 902118854 (<.01%)
cycles in affected programs: 419132 -> 419205 (0.02%)
helped: 4 / HURT: 2

DG2, Tiger Lake, Ice Lake, and Skylake had similar results (Ice Lake shown)
Totals:
Instrs: 152819500 -> 152819380 (-0.00%)
Cycles: 15014627187 -> 15014624437 (-0.00%)

Totals from 115 (0.02% of 662497) affected shaders:
Instrs: 28963 -> 28843 (-0.41%)
Cycles: 404582 -> 401832 (-0.68%)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19968>

16 months agonir/algebraic: Optimize some u2f of bfi
Ian Romanick [Tue, 26 Jul 2022 19:30:04 +0000 (12:30 -0700)]
nir/algebraic: Optimize some u2f of bfi

v2: Fix a copy-and-paste bug s/('find_lsb', a)/a/ in the patterns. See
piglit!819.

DG2, Tiger Lake, Ice Lake, Skylake, and Broadwell had similar results (Ice Lake shown)
total instructions in shared programs: 20570063 -> 20570033 (<.01%)
instructions in affected programs: 452 -> 422 (-6.64%)
helped: 30 / HURT: 0

total cycles in shared programs: 902118723 -> 902118781 (<.01%)
cycles in affected programs: 1762 -> 1820 (3.29%)
helped: 0 / HURT: 29

DG2, Tiger Lake, Ice Lake, and Skylake had similar results (Ice Lake shown)
Totals:
Instrs: 152819969 -> 152819500 (-0.00%)
Cycles: 15014628652 -> 15014627187 (-0.00%); split: -0.00%, +0.00%

Totals from 469 (0.07% of 662497) affected shaders:
Instrs: 7644 -> 7175 (-6.14%)
Cycles: 31787 -> 30322 (-4.61%); split: -4.90%, +0.29%

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19968>

16 months agointel/fs: Emit better code for bfi(..., 0)
Ian Romanick [Wed, 16 Nov 2022 21:12:50 +0000 (13:12 -0800)]
intel/fs: Emit better code for bfi(..., 0)

DG2, Tiger Lake, Ice Lake, and Skylake had similar results (Ice Lake shown)
total instructions in shared programs: 20570141 -> 20570063 (<.01%)
instructions in affected programs: 30679 -> 30601 (-0.25%)
helped: 77 / HURT: 0

total cycles in shared programs: 902113977 -> 902118723 (<.01%)
cycles in affected programs: 3255958 -> 3260704 (0.15%)
helped: 60 / HURT: 19

Broadwell
total instructions in shared programs: 18524633 -> 18524547 (<.01%)
instructions in affected programs: 34095 -> 34009 (-0.25%)
helped: 75 / HURT: 2

total cycles in shared programs: 949532394 -> 949543761 (<.01%)
cycles in affected programs: 3419107 -> 3430474 (0.33%)
helped: 57 / HURT: 24

total spills in shared programs: 22484 -> 22484 (0.00%)
spills in affected programs: 516 -> 516 (0.00%)
helped: 2 / HURT: 2

total fills in shared programs: 29346 -> 29338 (-0.03%)
fills in affected programs: 572 -> 564 (-1.40%)
helped: 4 / HURT: 0

Haswell
total instructions in shared programs: 17331356 -> 17331523 (<.01%)
instructions in affected programs: 27920 -> 28087 (0.60%)
helped: 41 / HURT: 4

total cycles in shared programs: 936603192 -> 936574664 (<.01%)
cycles in affected programs: 3417695 -> 3389167 (-0.83%)
helped: 28 / HURT: 21

total spills in shared programs: 19718 -> 19756 (0.19%)
spills in affected programs: 436 -> 474 (8.72%)
helped: 0 / HURT: 4

total fills in shared programs: 22547 -> 22607 (0.27%)
fills in affected programs: 444 -> 504 (13.51%)
helped: 0 / HURT: 4

Ivy Bridge
total cycles in shared programs: 463451277 -> 463451273 (<.01%)
cycles in affected programs: 95870 -> 95866 (<.01%)
helped: 3 / HURT: 2

DG2, Tiger Lake, Ice Lake, and Skylake had similar results (Ice Lake shown)
Totals:
Instrs: 152825278 -> 152819969 (-0.00%); split: -0.00%, +0.00%
Cycles: 15014075626 -> 15014628652 (+0.00%); split: -0.01%, +0.01%
Subgroup size: 8528536 -> 8528560 (+0.00%)
Send messages: 7711431 -> 7711464 (+0.00%)
Spill count: 99907 -> 99509 (-0.40%); split: -0.40%, +0.00%
Fill count: 202459 -> 201598 (-0.43%); split: -0.43%, +0.00%
Scratch Memory Size: 4376576 -> 4371456 (-0.12%)

Totals from 2915 (0.44% of 662497) affected shaders:
Instrs: 2288842 -> 2283533 (-0.23%); split: -0.24%, +0.01%
Cycles: 471633295 -> 472186321 (+0.12%); split: -0.27%, +0.39%
Subgroup size: 27488 -> 27512 (+0.09%)
Send messages: 151344 -> 151377 (+0.02%)
Spill count: 48091 -> 47693 (-0.83%); split: -0.83%, +0.00%
Fill count: 59053 -> 58192 (-1.46%); split: -1.46%, +0.00%
Scratch Memory Size: 1827840 -> 1822720 (-0.28%)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19968>

16 months agonir/algebraic: Lower some bfi with two constant sources
Ian Romanick [Wed, 16 Nov 2022 21:16:25 +0000 (13:16 -0800)]
nir/algebraic: Lower some bfi with two constant sources

All Haswell and newer Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 19907054 -> 19906882 (<.01%)
instructions in affected programs: 8103 -> 7931 (-2.12%)
helped: 52 / HURT: 0

total cycles in shared programs: 855779334 -> 855781791 (<.01%)
cycles in affected programs: 724201 -> 726658 (0.34%)
helped: 38 / HURT: 7

total sends in shared programs: 1039308 -> 1039302 (<.01%)
sends in affected programs: 162 -> 156 (-3.70%)
helped: 2 / HURT: 0

No shader-db changes on any older Intel platforms.

All Intel platforms had similar restuls. (Ice Lake shown)
Totals:
Instrs: 153117340 -> 152825222 (-0.19%); split: -0.19%, +0.00%
Cycles: 15011904351 -> 15014072944 (+0.01%); split: -0.04%, +0.05%
Send messages: 7711509 -> 7711421 (-0.00%)
Spill count: 100745 -> 99907 (-0.83%); split: -0.85%, +0.02%
Fill count: 203684 -> 202459 (-0.60%); split: -0.62%, +0.02%
Scratch Memory Size: 4403200 -> 4376576 (-0.60%)

Totals from 18603 (2.81% of 662496) affected shaders:
Instrs: 5258303 -> 4966185 (-5.56%); split: -5.56%, +0.00%
Cycles: 447391388 -> 449559981 (+0.48%); split: -1.29%, +1.77%
Send messages: 559231 -> 559143 (-0.02%)
Spill count: 5009 -> 4171 (-16.73%); split: -17.17%, +0.44%
Fill count: 8769 -> 7544 (-13.97%); split: -14.33%, +0.36%
Scratch Memory Size: 194560 -> 167936 (-13.68%)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19968>

16 months agointel/fs: Use nir_opt_reassociate_bfi
Ian Romanick [Tue, 26 Jul 2022 15:55:45 +0000 (08:55 -0700)]
intel/fs: Use nir_opt_reassociate_bfi

All Skylake and newer Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 19907072 -> 19907054 (<.01%)
instructions in affected programs: 8859 -> 8841 (-0.20%)
helped: 9 / HURT: 0

total cycles in shared programs: 855791238 -> 855779334 (<.01%)
cycles in affected programs: 3308294 -> 3296390 (-0.36%)
helped: 12 / HURT: 13

Broadwell
total instructions in shared programs: 17818231 -> 17817440 (<.01%)
instructions in affected programs: 9887 -> 9096 (-8.00%)
helped: 9 / HURT: 0

total cycles in shared programs: 902970035 -> 902941221 (<.01%)
cycles in affected programs: 2767243 -> 2738429 (-1.04%)
helped: 14 / HURT: 5

total spills in shared programs: 17784 -> 17718 (-0.37%)
spills in affected programs: 318 -> 252 (-20.75%)
helped: 1 / HURT: 0

total fills in shared programs: 25458 -> 24949 (-2.00%)
fills in affected programs: 1346 -> 837 (-37.82%)
helped: 1 / HURT: 0

Haswell
total instructions in shared programs: 16707799 -> 16707586 (<.01%)
instructions in affected programs: 24049 -> 23836 (-0.89%)
helped: 41 / HURT: 0

total cycles in shared programs: 882730648 -> 882723174 (<.01%)
cycles in affected programs: 5096737 -> 5089263 (-0.15%)
helped: 25 / HURT: 12

total spills in shared programs: 14937 -> 14909 (-0.19%)
spills in affected programs: 436 -> 408 (-6.42%)
helped: 4 / HURT: 0

total fills in shared programs: 17569 -> 17529 (-0.23%)
fills in affected programs: 444 -> 404 (-9.01%)
helped: 4 / HURT: 0

No shader-db changes on any older Intel platforms.

All Intel platforms had similar results. (Ice Lake shown)
Totals:
Instrs: 153118594 -> 153117340 (-0.00%); split: -0.00%, +0.00%
Cycles: 15011967556 -> 15011904351 (-0.00%); split: -0.00%, +0.00%
Fill count: 203692 -> 203684 (-0.00%)

Totals from 703 (0.11% of 662496) affected shaders:
Instrs: 192826 -> 191572 (-0.65%); split: -0.65%, +0.00%
Cycles: 29937640 -> 29874435 (-0.21%); split: -0.25%, +0.04%
Fill count: 4146 -> 4138 (-0.19%)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19968>

16 months agonir: Add optimization pass to reassociate some bfi instructions
Ian Romanick [Tue, 26 Jul 2022 15:49:32 +0000 (08:49 -0700)]
nir: Add optimization pass to reassociate some bfi instructions

The needs of this pass are ever so slightly more than what
nir_opt_algebraic can do. :( Specifically, it needs to be able to look
at the relationship of constant values used in an expression tree.

v2: Add nir_mov_alu to handle swizzles on the original sources.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19968>

16 months agozink: add some ci flakes
Mike Blumenkrantz [Wed, 14 Jun 2023 16:53:11 +0000 (12:53 -0400)]
zink: add some ci flakes

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23654>

16 months agoCI: Re-enable freedreno CI
Daniel Stone [Wed, 14 Jun 2023 15:46:37 +0000 (16:46 +0100)]
CI: Re-enable freedreno CI

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23108>

16 months agoci: Extend a618_vk_full runtime
Daniel Stone [Thu, 18 May 2023 14:43:47 +0000 (15:43 +0100)]
ci: Extend a618_vk_full runtime

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23108>

16 months agoci: Don't retry manual or scheduled jobs
Daniel Stone [Thu, 18 May 2023 14:37:00 +0000 (15:37 +0100)]
ci: Don't retry manual or scheduled jobs

Only retry when there's some kind of non-job failure, such as
runner-internal issues, or API/network issues, etc. If the job itself
fails or times out, then given the length of these jobs, there's no
point trying again and just tying up the job slots for even more hours.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23108>

16 months agoci: Elaborate causes for job retries
Daniel Stone [Thu, 18 May 2023 14:32:35 +0000 (15:32 +0100)]
ci: Elaborate causes for job retries

Rather than always retrying, only retry jobs on a limited set of causes.
This notably excludes retries when a job is stuck due to lack of runners
to schedule it; if we can't get a slot on a runner in time, there's no
reason to try again, since our window of opportunity has gone.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23108>

16 months agoci: Drop some skips of GL CTS ArraysOfArrays tests.
Emma Anholt [Mon, 5 Jun 2023 20:42:55 +0000 (13:42 -0700)]
ci: Drop some skips of GL CTS ArraysOfArrays tests.

My hope is that with my CTS fix, we can complete these all in time now.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23610>

16 months agoci: Drop skips for some previously-invalid CTS tests.
Emma Anholt [Mon, 5 Jun 2023 20:07:01 +0000 (13:07 -0700)]
ci: Drop skips for some previously-invalid CTS tests.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23610>

16 months agoci: Update to vulkan-cts-1.3.5.2 (and pull in some more fixes).
Emma Anholt [Mon, 5 Jun 2023 20:04:00 +0000 (13:04 -0700)]
ci: Update to vulkan-cts-1.3.5.2 (and pull in some more fixes).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23610>

16 months agoci/zink: Update current xfails on tgl.
Emma Anholt [Tue, 13 Jun 2023 00:14:11 +0000 (17:14 -0700)]
ci/zink: Update current xfails on tgl.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23610>

16 months agointel: Reduce cost of resetting last_grf_write.
Emma Anholt [Tue, 13 Jun 2023 21:18:28 +0000 (14:18 -0700)]
intel: Reduce cost of resetting last_grf_write.

In zink-on-anv fs-mod-dvec3-dvec3.shader_test, we were memsetting 2MB of
last_grf_write 2400 times, multiple times through the scheduler.  Just
resetting for the processed instructions reduces runtime from 21s to 16s.
No change on steam shader-db runtime across several runs.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23635>

16 months agointel: Allocate the last_grf_write once per scheduler.
Emma Anholt [Tue, 13 Jun 2023 20:46:56 +0000 (13:46 -0700)]
intel: Allocate the last_grf_write once per scheduler.

No need to re-calloc it per block when we're going to use it again.  Also,
this fixes the vec4 backend to avoid allocating giant grf_count-sized
arrays on the stack.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23635>

16 months agointel: Count reads_remaining across all blocks.
Emma Anholt [Tue, 13 Jun 2023 19:50:02 +0000 (12:50 -0700)]
intel: Count reads_remaining across all blocks.

We were zeroing it out per block, but it doesn't actually help to count
per block, since the question is "will scheduling this instruction free
the reg?".  Saves some memsetting, which was showing up high in the
profile (but not from this source).

No change on iris SKL shader-db.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23635>

16 months agoegl/dri2: trigger drawable invalidation from surface queries for zink
Mike Blumenkrantz [Thu, 4 May 2023 17:18:50 +0000 (13:18 -0400)]
egl/dri2: trigger drawable invalidation from surface queries for zink

this mimics dri3 behavior and avoids scenarios where renderbuffers can
get out of sync with their resources

fixes #6744

Acked-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22858>

16 months agolavapipe: add version uuid to shader binary validation
Mike Blumenkrantz [Tue, 13 Jun 2023 22:20:42 +0000 (18:20 -0400)]
lavapipe: add version uuid to shader binary validation

this ensures compatible shader binaries across versions

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23636>

16 months agor600: Disable SB if we use the ariable length DOT
Gert Wollny [Wed, 14 Jun 2023 12:33:25 +0000 (14:33 +0200)]
r600: Disable SB if we use the ariable length DOT

sb doesn't know about this instruction, so don't try to run the
optimizer.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23647>

16 months ago600/sfn: Trigger use of ACK for some barriers
Gert Wollny [Tue, 13 Jun 2023 15:13:07 +0000 (17:13 +0200)]
600/sfn: Trigger use of ACK for some barriers

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23647>

16 months agor600/sfn: move kill handling to fully scheduling
Gert Wollny [Sun, 14 May 2023 12:48:51 +0000 (14:48 +0200)]
r600/sfn: move kill handling to fully scheduling

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23647>

16 months agor600: fix handling of use_sb flag
Gert Wollny [Sun, 14 May 2023 12:50:24 +0000 (14:50 +0200)]
r600: fix handling of use_sb flag

The compiler will use the unsigned bit pattern of the check and combine this
with the 1 bit, which will always result in use_sb to be zero.

Fix this by making use_sb a bool

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23647>

16 months agozink: add a dgc debug mode for testing
Mike Blumenkrantz [Wed, 31 May 2023 15:20:17 +0000 (11:20 -0400)]
zink: add a dgc debug mode for testing

this is useful for drivers trying to implement DGC since there is no cts

do not use.

it will not make anything faster.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23550>

16 months agointel/fs: handle load_global_constant_uniform_block_intel
Lionel Landwerlin [Thu, 8 Jun 2023 18:53:02 +0000 (21:53 +0300)]
intel/fs: handle load_global_constant_uniform_block_intel

Again, load the data just once in GRF, share it across lanes.

Shader-db on dg2:

total instructions in shared programs: 23214555 -> 23215400 (<.01%)
instructions in affected programs: 199977 -> 200822 (0.42%)
helped: 3
HURT: 38
helped stats (abs) min: 5 max: 670 x̄: 283.67 x̃: 176
helped stats (rel) min: 1.34% max: 49.41% x̄: 22.15% x̃: 15.70%
HURT stats (abs)   min: 1 max: 185 x̄: 44.63 x̃: 32
HURT stats (rel)   min: 0.13% max: 42.86% x̄: 10.25% x̃: 9.30%
95% mean confidence interval for instructions value: -18.65 59.87
95% mean confidence interval for instructions %-change: 3.29% 12.47%
Inconclusive result (value mean confidence interval includes 0).

total loops in shared programs: 5928 -> 5928 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0

total cycles in shared programs: 851137495 -> 851152449 (<.01%)
cycles in affected programs: 16406137 -> 16421091 (0.09%)
helped: 9
HURT: 32
helped stats (abs) min: 10 max: 13498 x̄: 6443.22 x̃: 5581
helped stats (rel) min: 0.11% max: 4.75% x̄: 1.45% x̃: 0.34%
HURT stats (abs)   min: 3 max: 15056 x̄: 2279.47 x̃: 735
HURT stats (rel)   min: 0.10% max: 23.71% x̄: 4.58% x̃: 4.65%
95% mean confidence interval for cycles value: -1315.40 2044.87
95% mean confidence interval for cycles %-change: 1.71% 4.80%
Inconclusive result (value mean confidence interval includes 0).

total spills in shared programs: 11856 -> 11825 (-0.26%)
spills in affected programs: 2368 -> 2337 (-1.31%)
helped: 4
HURT: 0

total fills in shared programs: 16258 -> 16207 (-0.31%)
fills in affected programs: 2930 -> 2879 (-1.74%)
helped: 4
HURT: 0

total sends in shared programs: 1038194 -> 1038185 (<.01%)
sends in affected programs: 40 -> 31 (-22.50%)
helped: 4
HURT: 0
helped stats (abs) min: 1 max: 4 x̄: 2.25 x̃: 2
helped stats (rel) min: 10.00% max: 33.33% x̄: 21.46% x̃: 21.25%
95% mean confidence interval for sends value: -4.64 0.14
95% mean confidence interval for sends %-change: -40.41% -2.51%
Inconclusive result (value mean confidence interval includes 0).

LOST:   0
GAINED: 0

Some VK/DX titles result (on DG2 only), it's mostly additional
instruction counts except for the unity spaceship demo where a CS
shader gets additional SIMDness. The reason for additional
instructions is that since we're doing block loads, we need to find
the live channels in control flow to select a single lane value that
is valid.

aztec_ruins_high:
Totals from 3 (1.12% of 269) affected shaders:
Instrs: 17732 -> 17896 (+0.92%)
Cycles: 796518 -> 819302 (+2.86%)

cyberpunk_2077:
Totals from 17 (0.17% of 10301) affected shaders:
Instrs: 10848 -> 11658 (+7.47%)
Cycles: 248243 -> 259168 (+4.40%); split: -0.57%, +4.97%

fallout_4_dxvk_g2:
Totals from 2 (0.12% of 1638) affected shaders:
Instrs: 3157 -> 3368 (+6.68%)
Cycles: 487807 -> 490426 (+0.54%); split: -0.26%, +0.79%
Max live registers: 139 -> 141 (+1.44%)

red_dead_redemption2:
Totals from 68 (1.14% of 5970) affected shaders:
Instrs: 34871 -> 36486 (+4.63%)
Cycles: 551430 -> 565211 (+2.50%)
Send messages: 2074 -> 2072 (-0.10%)
Max live registers: 5078 -> 5077 (-0.02%)

total_war_warhammer2:
Totals from 5 (1.05% of 478) affected shaders:
Instrs: 6905 -> 6971 (+0.96%); split: -0.16%, +1.12%
Cycles: 97035 -> 97989 (+0.98%); split: -0.07%, +1.05%

unity spaceship demo (instruction count going up due to a CS shader
                      bump from SIMD8->16):
Totals from 53 (9.71% of 546) affected shaders:
Instrs: 223748 -> 233223 (+4.23%); split: -0.01%, +4.25%
Cycles: 23134697 -> 25207080 (+8.96%); split: -0.17%, +9.13%
Subgroup size: 480 -> 488 (+1.67%)
Spill count: 2156 -> 2242 (+3.99%); split: -0.19%, +4.17%
Fill count: 4617 -> 4845 (+4.94%); split: -0.09%, +5.02%
Max live registers: 5991 -> 6050 (+0.98%); split: -0.40%, +1.39%
Max dispatch width: 480 -> 488 (+1.67%)

witcher_3_dxvk_g2:
Totals from 27 (2.51% of 1074) affected shaders:
Instrs: 57067 -> 57677 (+1.07%); split: -0.03%, +1.10%
Cycles: 1397871 -> 1436704 (+2.78%); split: -0.35%, +3.13%

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23477>

16 months agonir: add a load_global_constant uniform intel variant
Lionel Landwerlin [Fri, 9 Jun 2023 10:57:43 +0000 (13:57 +0300)]
nir: add a load_global_constant uniform intel variant

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23477>

16 months agointel/fs: make use of load_ubo_uniform_block_intel
Lionel Landwerlin [Tue, 6 Jun 2023 15:03:26 +0000 (18:03 +0300)]
intel/fs: make use of load_ubo_uniform_block_intel

The principle is the same as the load_ssbo_uniform_block_intel.
Whenever we see a uniform offset, load the data only once in GRFs to
reduce register pressure.

Iris shader-db run on DG2 :

total instructions in shared programs: 23001325 -> 23094969 (0.41%)
instructions in affected programs: 1775989 -> 1869633 (5.27%)
helped: 764
HURT: 2097
helped stats (abs) min: 1 max: 102 x̄: 6.96 x̃: 2
helped stats (rel) min: 0.03% max: 16.91% x̄: 1.36% x̃: 0.63%
HURT stats (abs)   min: 1 max: 2461 x̄: 47.19 x̃: 7
HURT stats (rel)   min: <.01% max: 199.34% x̄: 5.91% x̃: 2.60%
95% mean confidence interval for instructions value: 25.43 40.03
95% mean confidence interval for instructions %-change: 3.60% 4.33%
Instructions are HURT.

total loops in shared programs: 5847 -> 5847 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0

total cycles in shared programs: 839329852 -> 845491482 (0.73%)
cycles in affected programs: 130229434 -> 136391064 (4.73%)
helped: 1098
HURT: 2228
helped stats (abs) min: 1 max: 130102 x̄: 1340.64 x̃: 22
helped stats (rel) min: <.01% max: 64.25% x̄: 4.03% x̃: 0.71%
HURT stats (abs)   min: 1 max: 185309 x̄: 3426.24 x̃: 87
HURT stats (rel)   min: <.01% max: 92.85% x̄: 8.12% x̃: 3.82%
95% mean confidence interval for cycles value: 1342.16 2362.97
95% mean confidence interval for cycles %-change: 3.70% 4.52%
Cycles are HURT.

total spills in shared programs: 10768 -> 11856 (10.10%)
spills in affected programs: 9717 -> 10805 (11.20%)
helped: 25
HURT: 28

total fills in shared programs: 13720 -> 16258 (18.50%)
fills in affected programs: 12016 -> 14554 (21.12%)
helped: 25
HURT: 28

total sends in shared programs: 1034790 -> 1031266 (-0.34%)
sends in affected programs: 33416 -> 29892 (-10.55%)
helped: 1005
HURT: 0
helped stats (abs) min: 1 max: 22 x̄: 3.51 x̃: 3
helped stats (rel) min: 1.69% max: 60.00% x̄: 15.20% x̃: 14.08%
95% mean confidence interval for sends value: -3.72 -3.29
95% mean confidence interval for sends %-change: -15.82% -14.57%
Sends are helped.

LOST:   26
GAINED: 183

shader-db on a number of VK/DX titles on DG2 :

 PERCENTAGE DELTAS  Shaders   Instrs    Cycles
 age_of_wonders_III 1928      +0.02%    -0.19%

 PERCENTAGE DELTAS       Shaders   Instrs    Cycles  Subgroup size Send messages Spill count Fill count Max live registers Max dispatch width
 assassins_creed_odyssey 2119      +1.12%    -0.42%      -0.03%        -0.29%       -9.10%     -4.26%         -0.64%             +0.65%

 PERCENTAGE DELTAS Shaders   Instrs    Cycles  Spill count Fill count Max live registers
 aztec_ruins_high  269       -0.05%    -0.45%     -0.29%     -7.27%         -0.33%

 PERCENTAGE DELTAS    Shaders   Instrs    Cycles  Max live registers Max dispatch width
 dark_souls_3_dxvk_g2 1420      +0.09%    +0.24%        +0.21%             +0.12%

(stats look bad, but it's just one shader affected)
 PERCENTAGE DELTAS Shaders   Instrs    Cycles  Spill count Fill count Scratch Memory Size Max live registers
 fallout_4_dxvk_g2 1638      +0.67%    +8.32%    +16.02%     +7.17%         +100.00%            +0.48%

 PERCENTAGE DELTAS    Shaders   Instrs    Cycles  Send messages Spill count Fill count Max live registers Max dispatch width
 red_dead_redemption2 5969      +0.16%    -0.04%      -0.04%       +0.01%     +0.05%         -0.20%             +0.04%

 PERCENTAGE DELTAS          Shaders   Instrs    Cycles  Send messages Max live registers Max dispatch width
 rise_of_the_tomb_raider_g2 12129     +2.19%    +1.36%      -1.23%          -0.36%             +2.04%

 PERCENTAGE DELTAS Shaders   Instrs    Cycles  Send messages Max live registers
 shooter-game      693       +0.07%    -0.89%      -0.09%          -0.09%

 PERCENTAGE DELTAS Shaders   Instrs    Cycles  Send messages Max live registers Max dispatch width
 talos_g2          1140      +0.37%    +3.80%      -0.86%          -0.67%             +0.19%

 PERCENTAGE DELTAS    Shaders   Instrs    Cycles  Max live registers Max dispatch width
 total_war_warhammer2 477       +0.25%    +0.66%        -0.17%             +0.10%

 PERCENTAGE DELTAS Shaders   Instrs    Cycles  Send messages Max live registers Max dispatch width
 witcher_3_dxvk_g2 1074      +0.75%   -10.45%      -0.15%          -0.16%             -0.16%

 PERCENTAGE DELTAS      Shaders   Instrs    Cycles  Send messages Max live registers
 wolfenstein_youngblood 1111      +0.52%    +0.66%      -0.59%          -0.03%

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23477>

16 months agonir: add a new ubo uniform loading intrinsic for intel
Lionel Landwerlin [Tue, 6 Jun 2023 15:02:42 +0000 (18:02 +0300)]
nir: add a new ubo uniform loading intrinsic for intel

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23477>

16 months agointel/fs: avoid reusing the VGRF for uniform load_ubo
Lionel Landwerlin [Tue, 13 Jun 2023 17:37:22 +0000 (20:37 +0300)]
intel/fs: avoid reusing the VGRF for uniform load_ubo

Only found 3 shaders affected in Red Dead Redemption :

Totals from 3 (0.05% of 5969) affected shaders:
Instrs: 2246 -> 2230 (-0.71%)
Cycles: 156506 -> 148402 (-5.18%); split: -5.23%, +0.05%

This will have a larger effect when we add the
load_ubo_uniform_block_intel intrinsic where we will have larger
blocks (vec8/vec16 vs vec4 only now).

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23477>

16 months agointel/fs: print identation for control flow
Lionel Landwerlin [Fri, 9 Jun 2023 08:48:26 +0000 (11:48 +0300)]
intel/fs: print identation for control flow

INTEL_DEBUG=optimizer output changes from :

{ 10}   40: cmp.nz.f0.0(8) null:F, vgrf3470:F, 0f
{ 10}   41: (+f0.0) if(8) (null):UD,
{ 11}   42: txf_logical(8) vgrf3473:UD, vgrf250:D(null):UD, 0d(null):UD(null):UD(null):UD(null):UD, 31u, 0u(null):UD(null):UD(null):UD, 3d, 0d
{ 12}   43: and(8) vgrf262:UD, vgrf3473:UD, 2u
{ 11}   44: cmp.nz.f0.0(8) null:D, vgrf262:D, 0d
{ 10}   45: (+f0.0) if(8) (null):UD,
{ 11}   46: mov(8) vgrf270:D, -1082130432d
{ 12}   47: mov(8) vgrf271:D, 1082130432d
{ 14}   48: mov(8) vgrf274+0.0:D, 0d
{ 14}   49: mov(8) vgrf274+1.0:D, 0d

to :

{ 10}   40: cmp.nz.f0.0(8) null:F, vgrf3470:F, 0f
{ 10}   41: (+f0.0) if(8) (null):UD,
{ 11}   42:   txf_logical(8) vgrf3473:UD, vgrf250:D(null):UD, 0d(null):UD(null):UD(null):UD(null):UD, 31u, 0u(null):UD(null):UD(null):UD, 3d, 0d
{ 12}   43:   and(8) vgrf262:UD, vgrf3473:UD, 2u
{ 11}   44:   cmp.nz.f0.0(8) null:D, vgrf262:D, 0d
{ 10}   45:   (+f0.0) if(8) (null):UD,
{ 11}   46:     mov(8) vgrf270:D, -1082130432d
{ 12}   47:     mov(8) vgrf271:D, 1082130432d
{ 14}   48:     mov(8) vgrf274+0.0:D, 0d
{ 14}   49:     mov(8) vgrf274+1.0:D, 0d

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23477>

16 months agorusticl/queue: overhaul of the queue+event handling
Karol Herbst [Tue, 13 Jun 2023 00:07:02 +0000 (02:07 +0200)]
rusticl/queue: overhaul of the queue+event handling

This new approach handles things as follows:
1. Fences won't be attached to events anymore, applications only wait on
   the cv attached to the event.
2. Only the queue is allowed to update event status for non user events.
   This will eliminate all remaining status updating races between the
   queue and applications waiting on events.
3. Queue minimized flushing by bundling events
4. Increase cv wait timeout as there is really no point in waking up too
   often.

Reduces amount of emited fences on radeonsi in luxmark 3.1 luxball by 90%

Signed-off-by: Karol Herbst <git@karolherbst.de>
Reviewed by Nora Allen <blackcatgames@protonmail.com>

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23612>

16 months agobroadcom/compiler: only use last thread switch flag to detect final section
Iago Toral Quiroga [Thu, 13 Apr 2023 08:31:17 +0000 (10:31 +0200)]
broadcom/compiler: only use last thread switch flag to detect final section

Since commit 'c98ddc778a3 broadcom/compiler: force a last thrsw for spilling'
we always ensure we signal the last thread section explicitly with a
last thread switch.

Relying on VPM stores to detect the last thread section is particularly bad,
because we can have VPM stores occurring quite early in a shader program,
which would disable TMU spilling almost entirely.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22461>

16 months agobroadcom/compiler: clarify use of QFILE_VPM
Alejandro Piñeiro [Fri, 12 May 2023 12:04:56 +0000 (14:04 +0200)]
broadcom/compiler: clarify use of QFILE_VPM

This was only used for version < 40 (See commit 22a02f3e3).

Adding some extra explanations and asserts of places where it is used.

As we are here also move the definition of a register with QFILE_VPM,
to avoid defining it if not needed.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22984>

16 months agointel/fs: fix bindless/shared surface mistake
Lionel Landwerlin [Thu, 8 Jun 2023 22:05:34 +0000 (01:05 +0300)]
intel/fs: fix bindless/shared surface mistake

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 068bf1378d ("intel/fs: enable SSBO accesses through the bindless heap")
Tested-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23536>

16 months agoanv: enable CmdCopyQueryPoolResults to use shader for copies
Lionel Landwerlin [Tue, 16 May 2023 19:11:47 +0000 (22:11 +0300)]
anv: enable CmdCopyQueryPoolResults to use shader for copies

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agointel/ds: add query count in query tracepoints
Lionel Landwerlin [Tue, 13 Jun 2023 08:11:27 +0000 (11:11 +0300)]
intel/ds: add query count in query tracepoints

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: add shaders for copying query results
Lionel Landwerlin [Tue, 16 May 2023 18:59:33 +0000 (21:59 +0300)]
anv: add shaders for copying query results

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: generalize internal kernel concept
Lionel Landwerlin [Tue, 16 May 2023 13:10:31 +0000 (16:10 +0300)]
anv: generalize internal kernel concept

We'll add more of those kernels for other purposes.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: add support for simple internal compute shaders
Lionel Landwerlin [Tue, 16 May 2023 10:13:41 +0000 (13:13 +0300)]
anv: add support for simple internal compute shaders

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: factor out generation kernel dispatch into helper
Lionel Landwerlin [Tue, 16 May 2023 09:54:39 +0000 (12:54 +0300)]
anv: factor out generation kernel dispatch into helper

We would like to reuse this mechanism to dispatch different types of
internal shader. Those would replace some of the command streamer
commands we currently use.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: limit ANV_PIPE_RENDER_TARGET_BUFFER_WRITES to blorp operations using 3D
Lionel Landwerlin [Wed, 17 May 2023 10:40:14 +0000 (13:40 +0300)]
anv: limit ANV_PIPE_RENDER_TARGET_BUFFER_WRITES to blorp operations using 3D

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: fix incorrect batch for 3DSTATE_CONSTANT_ALL emission
Lionel Landwerlin [Tue, 16 May 2023 06:53:42 +0000 (09:53 +0300)]
anv: fix incorrect batch for 3DSTATE_CONSTANT_ALL emission

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: c950fe97a0 ("anv: implement generated (indexed) indirect draws")
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: disable mesh/task for generated draws
Lionel Landwerlin [Tue, 16 May 2023 06:20:09 +0000 (09:20 +0300)]
anv: disable mesh/task for generated draws

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: c950fe97a0 ("anv: implement generated (indexed) indirect draws")
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoanv: only disable mesh when enabled at the VkDevice level
Lionel Landwerlin [Tue, 16 May 2023 06:25:17 +0000 (09:25 +0300)]
anv: only disable mesh when enabled at the VkDevice level

Saving ourselves some instructions since it's not going to get used.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23074>

16 months agoradv: fix smooth lines with graphics pipeline library
Samuel Pitoiset [Wed, 24 May 2023 06:20:33 +0000 (08:20 +0200)]
radv: fix smooth lines with graphics pipeline library

Need to be defensive here.

Fixes recent CTS
dEQP-VK.pipeline.fast_linked_library.extended_dynamic_state.*.line_raster_mode_smooth.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23200>

16 months agoradv: do not force VRS 1x1 when smooth lines are enabled
Samuel Pitoiset [Wed, 24 May 2023 12:51:46 +0000 (14:51 +0200)]
radv: do not force VRS 1x1 when smooth lines are enabled

Otherwise this will break VRS with GPL.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23200>

16 months agoradv: add a helper for forcing VRS 1x1 in some situations
Samuel Pitoiset [Wed, 24 May 2023 12:44:54 +0000 (14:44 +0200)]
radv: add a helper for forcing VRS 1x1 in some situations

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23200>

16 months agoradv: gather info about load_poly_line_smooth_enabled
Samuel Pitoiset [Wed, 24 May 2023 12:42:00 +0000 (14:42 +0200)]
radv: gather info about load_poly_line_smooth_enabled

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23200>

16 months agoradv/video: reject unsupported hevc profiles and bit depths
Lynne [Tue, 13 Jun 2023 15:50:29 +0000 (17:50 +0200)]
radv/video: reject unsupported hevc profiles and bit depths

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23625>

16 months agoradv/video: reject non-8bit H264
Lynne [Tue, 13 Jun 2023 15:50:07 +0000 (17:50 +0200)]
radv/video: reject non-8bit H264

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23625>

16 months agoradv/video: reject general unsupported video formats
Lynne [Tue, 13 Jun 2023 15:40:50 +0000 (17:40 +0200)]
radv/video: reject general unsupported video formats

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23625>

16 months agosvga: fix shader type after ntt
Charmaine Lee [Wed, 14 Jun 2023 00:14:41 +0000 (03:14 +0300)]
svga: fix shader type after ntt

Set shader type to TGSI after ntt. Fixes assert in svga_create_shader.

Fixes: 0ac95418048 ("gallium: Drop PIPE_SHADER_CAP_PREFERRED_IR")

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23638>

16 months agozink: add some ntv asserts for ms txf
Mike Blumenkrantz [Mon, 8 May 2023 12:55:03 +0000 (08:55 -0400)]
zink: add some ntv asserts for ms txf

it's illegal to do ms txf without a ms image

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22900>

16 months agozink: assert that ntv image creation isn't clobbering existing images
Mike Blumenkrantz [Mon, 8 May 2023 12:54:26 +0000 (08:54 -0400)]
zink: assert that ntv image creation isn't clobbering existing images

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22900>

16 months agozink: massively shrink qbo size for timestamp queries
Mike Blumenkrantz [Tue, 13 Jun 2023 17:12:37 +0000 (13:12 -0400)]
zink: massively shrink qbo size for timestamp queries

timestamp queries can only ever record a single value, so
the qbo only needs to be large enough for that single value

fixes #9092

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23631>

16 months agoradv/video: program hevc max dec pic buffering correctly
Dave Airlie [Tue, 13 Jun 2023 00:44:33 +0000 (10:44 +1000)]
radv/video: program hevc max dec pic buffering correctly

This programs it like vaapi does.

Reviewed-by: Lynne <dev@lynne.ee>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23613>

16 months agoradv/video: program frame number correctly.
Dave Airlie [Mon, 12 Jun 2023 01:48:58 +0000 (11:48 +1000)]
radv/video: program frame number correctly.

doesn't appear to fix anything, but is more correct.

Reviewed-by: Lynne <dev@lynne.ee>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23613>

16 months agoanv: always report all pipeline stats regardless of stages
Lionel Landwerlin [Fri, 9 Jun 2023 13:52:17 +0000 (16:52 +0300)]
anv: always report all pipeline stats regardless of stages

Tools like the scripts in shader-db expect all the fields to be there,
as the stats are put into a CSV file. So just report 0 if a stage
doesn't support workgroup memory size.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23559>

16 months agoanv: report max simd width only once for fragment shaders
Lionel Landwerlin [Fri, 9 Jun 2023 13:32:17 +0000 (16:32 +0300)]
anv: report max simd width only once for fragment shaders

Reporting the value multiple times is confusing to shader-db scripts
because it believes multiple shaders are affected.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23559>

16 months agoanv: deal with unsupported VkImageFormatListCreateInfo::pViewFormats
Lionel Landwerlin [Mon, 12 Jun 2023 18:05:44 +0000 (21:05 +0300)]
anv: deal with unsupported VkImageFormatListCreateInfo::pViewFormats

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 697ed61e7c ("anv: Improve image/view usage bits verification")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9190
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23606>

16 months agoradv: Enable ray tracing pipelines by default
Friedrich Vock [Mon, 12 Jun 2023 17:50:55 +0000 (19:50 +0200)]
radv: Enable ray tracing pipelines by default

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23603>

16 months agoradv: Add RADV_DEBUG=nort
Friedrich Vock [Mon, 12 Jun 2023 17:41:18 +0000 (19:41 +0200)]
radv: Add RADV_DEBUG=nort

This makes vkCmdTraceRaysKHR and ray queries no-ops and is useful
for triaging GPU hangs with raytracing.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23603>

16 months agolavapipe: use PACKAGE_VERSION for cache uuid in release builds
Mike Blumenkrantz [Wed, 4 Jan 2023 15:22:57 +0000 (10:22 -0500)]
lavapipe: use PACKAGE_VERSION for cache uuid in release builds

MESA_GIT_SHA1 is defined as "", which results in invalid access

fixes #7673

Fixes: b38879f8c5f ("vallium: initial import of the vulkan frontend")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20520>

16 months agoradv: directly use vk_format_map for vertex input
Mike Blumenkrantz [Thu, 8 Jun 2023 14:17:26 +0000 (10:17 -0400)]
radv: directly use vk_format_map for vertex input

this is much faster than calling a non-inline function with
conditionals to index the same array

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23599>

16 months agovk: make vk_format_map[] public
Mike Blumenkrantz [Thu, 8 Jun 2023 14:16:54 +0000 (10:16 -0400)]
vk: make vk_format_map[] public

having to go through a function call for non-planar mappings can be
very slow

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23599>

16 months agoetnaviv: only emit sampler config for changed samplers
Lucas Stach [Sun, 11 Jun 2023 18:29:06 +0000 (20:29 +0200)]
etnaviv: only emit sampler config for changed samplers

Currently SAMPLER_CONFIG0 is always emitted to either update the active
configuration or disable the sampler. With NTE this always emits 32 state
dwords, while there are a lot of cases that only use a small number of
samplers and never change the other samplers from their disabled state.

Track the active samplers from the last emit, so we can skip the state
emission when the sampler is already disabled. Only emit the full state
after a context flush where we don't know the previous sampler state of
the GPU.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23579>

16 months agozink: only try to create srgb mutable images if the vk format is supported
Mike Blumenkrantz [Thu, 8 Jun 2023 01:21:04 +0000 (21:21 -0400)]
zink: only try to create srgb mutable images if the vk format is supported

otherwise this is just a regular single-format image

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23626>

16 months agoradv: tweak gfx pipeline stage binding
Mike Blumenkrantz [Wed, 7 Jun 2023 21:00:42 +0000 (17:00 -0400)]
radv: tweak gfx pipeline stage binding

iterating all the stages like this ends up unnecessarily calling
through to geometry stage binds when no shader was bound and no shader
is being bound by the power of optimization, so instead only do the unbind
part for the stages that are being unbound

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23600>

16 months agonir: Remove non-scoped barriers
Alyssa Rosenzweig [Tue, 6 Jun 2023 20:55:35 +0000 (16:55 -0400)]
nir: Remove non-scoped barriers

Nothing uses them anymore.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23191>