panfrost: Translate fixed-function blend at CSO create

Fixes: 93824b6451a ("panfrost: Move the blend logic out of the gallium driver")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

panfrost: Move blend properties to CSO create

Fixes: 93824b6451a ("panfrost: Move the blend logic out of the gallium driver")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

panfrost: Fix pan_blend_to_fixed_function_equation prototype

Doesn't need the whole state, just the equation itself.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Fixes: 93824b6451a ("panfrost: Move the blend logic out of the gallium driver")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

panfrost: Fix blend fixed-function prototype

Needs to be broken up into different functions for disjoint state.
Simpler prototypes and allows matching to CSOs.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

panfrost: Fix blend constant fetch prototype

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Fixes: 93824b6451a ("panfrost: Move the blend logic out of the gallium driver")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

panfrost: Fix is_opaque prototype

Fixes: 93824b6451a ("panfrost: Move the blend logic out of the gallium driver")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

panfrost: Fix the reads_dest prototype

Takes too much state, only pass what we need.

Fixes: 93824b6451a ("panfrost: Move the blend logic out of the gallium driver")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

panfrost: Fix blending for unbacked MRT

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10869>

aco: Implement bvh64_intersect_ray_amd intrinsic.

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10818>

nir: Add bvh64_intersect_ray_amd intrinsic.

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10818>

r600: Enable NIR debug flags also for Cayman

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: legalize image access on Cayman

This,  seems to avoid hardware resets with the
  ARB_shader_image_load_store-invalid
piglit.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Add lowering pass to legalize image access

Make sure only existing images are accessed and that the accessed
coordinates are within the image.
The generated code is quite exponsive, because it encapsulates each
access to an image with two if statements, one to check whether the
image index actually exists (this will get optimized away if the
image selection is direct), and one if statement to check whether
the coordinates are in range. For that reason it will only be enabled
for Cayman were invalid access seems to raise more issues.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Fix cube query layer number for indirect access

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: read number of images from shader info

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Fix Geometry shader for Cayman

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Fix Cayman SSBO write with more than one value

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Fix ssbo/image atomic swap for Cayman

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Fix texture gather for Cayman

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Use unified code path for index register load

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Use unified index register code for samplers

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Fix Cayman trans ops

v2: return from emit_sin_r600 on > Cayman (Kenn Glennard)

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: count only distinct literals per instruction group

This is to avoid useless instruction group splits.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Don't check the faction when searching for the input slot

All loads from the same driver location are merged into one variable,
so there is no need to check the frachtion.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Lower FS pos input w-transform in NIR

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: force new CF if fetch through TC would be used in same clause

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Use valid pixel mode for SSBO and Image result fetches

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600/sfn: Use valid pixel mode only in fragment shaders

The doc says that it should only be set in the FS.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

r600: don't put INTERP_X and INTERP_Z into one instruction group

Apparently this is not allowed and results in interpolation errors.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10608>

panfrost: Reduce blitter pool size

Blit shaders are small and the average app doesn't use many of them, so
try to pack in a single 4k BO. Saves 60k in a lot of simple apps.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

panfrost: Pool texture views

No need to stick these in separate BOs.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

panfrost: Pool shaders

Now we can do so without leaking memory :-)

Likewise use a pan_pool for RSDs, to share the common path.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

panfrost: Add reference type for unowned pool

This allows implementing the common pattern of allocating from an
unowned pool and immediately taking a reference.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

panfrost: Make pool slab size configurable

Different pools have different expected sizes.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

panfrost: Label pools

Allows the allocated BOs to be labeled more intelligently.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

panfrost: Label all BOs in userspace

Doesn't do any fancy cross-process labeling like @shadeslayer's patches
but helps with all your intra-process labeling needs.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

panfrost: Add unowned mode to pan_pool

I would like to reuse pan_pool for persistent uploads (shaders and CSOs)
in Gallium. In theory u_upload_mgr is more appropriate, but pan_pool is
already a knockoff u_upload_mgr, so might as well finish the job.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

pan/mdg: Set lower_uniforms_to_ubo

Rip off the band-aid. No other driver in Mesa has this combination, and
there's a reason for that. It confuses nir_to_tgsi; this commit fixes
GALLIUM_HUD (again) on Midgard.

shader-db stats aren't as bad as they appear at first blush, since the
added cycles are from added if-else branches (and only one side of the
if is taken on Midgard, which does no warping*).

total instructions in shared programs: 97036 -> 98107 (1.10%)
instructions in affected programs: 8297 -> 9368 (12.91%)
helped: 0
HURT: 45
HURT stats (abs)   min: 1 max: 52 x̄: 23.80 x̃: 25
HURT stats (rel)   min: 0.61% max: 61.90% x̄: 16.18% x̃: 14.66%
95% mean confidence interval for instructions value: 20.04 27.56
95% mean confidence interval for instructions %-change: 12.62% 19.74%
Instructions are HURT.

total bundles in shared programs: 45507 -> 46091 (1.28%)
bundles in affected programs: 3138 -> 3722 (18.61%)
helped: 2
HURT: 40
helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
helped stats (rel) min: 1.54% max: 2.17% x̄: 1.86% x̃: 1.86%
HURT stats (abs)   min: 2 max: 44 x̄: 14.65 x̃: 12
HURT stats (rel)   min: 2.70% max: 50.00% x̄: 23.03% x̃: 21.25%
95% mean confidence interval for bundles value: 10.35 17.46
95% mean confidence interval for bundles %-change: 16.81% 26.88%
Bundles are HURT.

total quadwords in shared programs: 76952 -> 77849 (1.17%)
quadwords in affected programs: 6556 -> 7453 (13.68%)
helped: 2
HURT: 44
helped stats (abs) min: 1 max: 5 x̄: 3.00 x̃: 3
helped stats (rel) min: 1.73% max: 4.55% x̄: 3.14% x̃: 3.14%
HURT stats (abs)   min: 2 max: 58 x̄: 20.52 x̃: 18
HURT stats (rel)   min: 2.11% max: 46.34% x̄: 17.20% x̃: 12.96%
95% mean confidence interval for quadwords value: 15.18 23.82
95% mean confidence interval for quadwords %-change: 12.68% 19.96%
Quadwords are HURT.

total registers in shared programs: 6966 -> 6925 (-0.59%)
registers in affected programs: 347 -> 306 (-11.82%)
helped: 26
HURT: 8
helped stats (abs) min: 1 max: 4 x̄: 2.04 x̃: 2
helped stats (rel) min: 6.67% max: 42.86% x̄: 20.92% x̃: 22.22%
HURT stats (abs)   min: 1 max: 5 x̄: 1.50 x̃: 1
HURT stats (rel)   min: 9.09% max: 50.00% x̄: 17.19% x̃: 11.81%
95% mean confidence interval for registers value: -1.85 -0.56
95% mean confidence interval for registers %-change: -18.97% -4.93%
Registers are helped.

total threads in shared programs: 5040 -> 5050 (0.20%)
threads in affected programs: 13 -> 23 (76.92%)
helped: 10
HURT: 1
helped stats (abs) min: 1 max: 2 x̄: 1.10 x̃: 1
helped stats (rel) min: 100.00% max: 100.00% x̄: 100.00% x̃: 100.00%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 50.00% max: 50.00% x̄: 50.00% x̃: 50.00%
95% mean confidence interval for threads value: 0.44 1.38
95% mean confidence interval for threads %-change: 55.98% 116.75%
Threads are helped.

Fixes: 24d7c413fe7 ("panfrost: Enable packed uniforms.")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

pan/mdg: Fix spills to TLS

LOCAL_STORAGE.zw is workgroup local memory, whereas LOCAL_STORAGE.xy is
thread local memory. Likewise PC_SP.zw is the stack pointer, which is
initialized to (LOCAL_STORAGE.zw + offset) but is modifiable by the
shader. Panfrost doesn't modify the s tack pointer, and the register
allocation logic assumes a zero offset, so let's always spill to thread
local memory = LOCAL_STORAGE.xy, as was intended by Italo's cleanup.

This is visible on any shader that spills. Compute shaders aren't
advertised yet, so WLS will be null, causing a fault like the following
(reproduced on Mali T860 with the glyphy trace):

[15634.148873] panfrost ff9a0000.gpu: Unhandled Page fault in AS0 at VA 0x0000000000000000
          Reason: TODO
          raw fault status: 0x70003C2
          decoded fault status: SLAVE FAULT
          exception type 0xC2: TRANSLATION_FAULT_LEVEL2
          access type 0x3: WRITE
          source id 0x700
[15634.658170] panfrost ff9a0000.gpu: gpu sched timeout, js=0, config=0x3300, status=0x8, head=0x31d4540,
tail=0x31d4540, sched_job=00000000e8101b2e

Fixes: 6a12ea02fee ("pan/mdg: properly encode/decode ldst instructions")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

pan/mdg: Fix output types for scalar fields

Already fixed vector, but scalar was missed.

Fixes: 4d9c0a32e7e ("pan/mdg: Use _output_ type for outmod printing")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

pan/mdg: Remove unused midgard_int_alu_op_prefix

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10866>

radv: Use correct border swizzle on GFX9+.

We only need the format swizzle, not the full swizzle.

Fixes: 57e796a12a8 ("radv: Implement VK_EXT_custom_border_color")
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4020
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9731>

virgl: implement EXT_multisampled_render_to_texture

Signed-off-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10500>

radv: Implement vkCmdTraceRays.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9767>

radv: Add RT pipeline bind.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9767>

radv: Add support for RT bind point.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9767>

radv: Use global BO list with raytracing.

The shader binding tables in vkCmdTraceRays have no buffer reference.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9767>

aco: Add load_sbt_amd intrinsic implementation.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9767>

radv: Add sbt descriptors user SGPR input.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9767>

nir: Add load_sbt_amd intrinsic.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9767>

ci/broadcom: update expected results

Fix also some typos in the expected failed results.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10865>

pps: Intel documentation

Add documentation for the Intel Perfetto datasource.

Signed-off-by: Antonio Caggiano <antonio.caggiano@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10216>

pps: Intel pps driver

Add the Intel pps driver using functionalities provided by
libintel_perf.

v2: Fix build with perfetto not enabled.
v3: Open perf stream with no filtering.
v4: Drop usage of inc/dec_n_users.
v5: Isolate intel_perf in its own class.

Signed-off-by: Antonio Caggiano <antonio.caggiano@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10216>

intel/perf: allow metric sets to be loaded with on OA reports

A bunch of performance counters rely on register snapshots on top of
the OA reports. Those are already conditional to the query mode in the
equations :

   availability="true $QueryMode &amp;&amp;"

This change allows to disable counters that are only available with
additional register snapshots. This will be useful if you only want to
OA reports to extract performance counter values.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Antonio Caggiano <antonio.caggiano@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10216>

intel/perf: allow opening perf stream with no context filtering

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Antonio Caggiano <antonio.caggiano@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10216>

intel/perf: Extern C

Allow libintel_perf to be included as a dependency from a C++ project by
wrapping some declaration within an extern "C" block, and then add a
function to allow direct reading of the OA stream.

v2: Don't expose internal helpers (Lionel)

Signed-off-by: Antonio Caggiano <antonio.caggiano@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10216>

docs/features: mark a few more extensions as done for zink

These are all done, but wasn't marked as done yet.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10861>

docs/features: document GL_ARB_ES3_2_compatibility support for zink

We support this extension on Zink even if we don't support ES 3.2,
because the compatibility extension doesn't require advanced blending.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10861>

panfrost/ci: Improve coverage for T860

By skipping slower tests such as `.*3_level_array.*`, we can add more
tests to CI without dramatically increase job duration.

Besides that, this patch also increases coverage by adding different
kinds of tests that were previously untested, failing or slowed CI too
much.

Signed-off-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10855>

aco: Eliminate useless exec writes in jump threading.

Eliminate exec writes which are unused by subsequent instructions.

Fossil DB results on Sienna Cichlid:

Totals from 80960 (54.03% of 149839) affected shaders:
CodeSize: 162953748 -> 161749372 (-0.74%)
Instrs: 31462273 -> 31161179 (-0.96%)
Copies: 2171239 -> 1942293 (-10.54%)
Branches: 807771 -> 807747 (-0.00%)

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10691>

aco: Refactor SSA elimination phi info to use vector instead of map.

No Fossil DB changes.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10691>

aco: Don't use s_and_saveexec with branches when exec is constant.

When exec is constant, we can remember the constant as the old exec,
and just copy the condition and use it as the new exec. There is no
need to save the constant.

Due to using p_parallelcopy which is lowered to s_mov_b64 (or 32),
many exec restores now become copies, hence the increase in the copy
stats.

Fossil DB changes on Sienna Cichlid:

Totals from 73969 (49.37% of 149839) affected shaders:
SpillSGPRs: 1768 -> 1610 (-8.94%)
CodeSize: 99053892 -> 99047884 (-0.01%); split: -0.02%, +0.01%
Instrs: 19372852 -> 19370398 (-0.01%); split: -0.02%, +0.01%
VClause: 515154 -> 515142 (-0.00%); split: -0.00%, +0.00%
SClause: 719236 -> 718395 (-0.12%); split: -0.14%, +0.02%
Copies: 1109770 -> 1254634 (+13.05%); split: -0.07%, +13.12%
Branches: 374338 -> 374348 (+0.00%); split: -0.00%, +0.00%
PreSGPRs: 1776481 -> 1653761 (-6.91%)

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10691>

aco: Remember when exec mask is const, and restore the const then.

Previously, we would store even the constant -1 exec mask from the
beginning of every merged shader. With this change it is no longer
necessary because we can restore to constant exec mask directly.

Hence, this frees up a register pair (single register for Wave32)
in every merged shader.

No Fossil DB changes.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10691>

aco: Use Operand instead of Temp for the exec mask stack.

This will enable us to store non-temporary values,
such as constant operands there.

No Fossil DB changes.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10691>

aco: Determine whether a few more instructions need exec.

These don't really need the exec mask (and never have), but we haven't
needed to include them in needs_exec_mask yet.

No Fossil DB changes.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10691>

broadcom/ci: update fail list for v3dv

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10801>

broadcom/compiler: consider RT component size when lowering logic ops in Vulkan

In Vulkan we configure our integer RTs to clamp automatically, so with logic
operations we need to be careful and avoid overflows by discarding any bits
that won't fit in the RT component size.

Fixes remaining CTS test failures in:
dEQP-VK.pipeline.logic_op.*

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10801>

broadcom/compiler: don't emit TLB loads for components that don't exist

This avoids debug builds to assert crash. Components that don't exist
won't be used and will be eventually DCEd, so simply lower them to 0.

Fixes CTS tests like these in debug builds:
dEQP-VK.pipeline.logic_op.r8_uint.clear
dEQP-VK.pipeline.logic_op.r8_uint.and
dEQP-VK.pipeline.logic_op.r8_uint.and_reverse
dEQP-VK.pipeline.logic_op.r8_uint.copy
dEQP-VK.pipeline.logic_op.r8_uint.and_inverted
dEQP-VK.pipeline.logic_op.r8_uint.no_op
dEQP-VK.pipeline.logic_op.r8_uint.xor
dEQP-VK.pipeline.logic_op.r8_uint.or

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10801>

i965: export GEM handle with RDWR access rights

There is a regression that made it impossible to export gem
handles with write access.

That is, a client may export gem handles of each buffer plane, then
export dmabuf fds using these handles, and mmap these dmabuf in
a different process (this is what Chromium does).

After https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4861,
it became impossible as mmap resulted in EACCESS error as slightly
different approach was taken for exporting these gem handles.

This CL fixes exporting gem handles (which are exported from dmabuf
fds) by adding the DRM_RDWR flag.

Cc: mesa-stable
Fixes #3119

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10850>

radv: add missing entrypoints for VK_EXT_extended_dynamic_state2

Fixes warnings with dEQP-VK.api.version_check.entry_points.

Fixes: d96507b73f2 ("radv: advertise VK_EXT_extended_dynamic_state2")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10831>

etnaviv: fix vertex sampler setup

The start offset of the vertex samplers isn't zero, but the indexing of
the passed in views array is still zero based. Use the correct indexing
variable to fix vertex sampler setup.

Cc: <mesa-stable@lists.freedesktop.org>
Fixes: 81ab9fe2d0c2 ("etnaviv: handle NULL views in set_sampler_views")
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10724>

iris: export GEM handle with RDWR access rights

There is a regression that made it impossible to export gem
handles with write access.

That is, a client may export gem handles of each buffer plane, then
export dmabuf fds using these handles, and mmap these dmabuf in
a different process (this is what Chromium does).

After https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4861,
it became impossible as mmap resulted in EACCESS error as slightly
different approach was taken for exporting these gem handles.

This CL fixes exporting gem handles (which are exported from dmabuf
fds) by adding the DRM_RDWR flag.

Cc: mesa-stable
Fixes #3119

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10851>

nir/opt_access: fix getting variables in presence of similar bindings/desc

It's perfectly legal to declare multiple SSBOs that point to the same
binding/descriptor_set with different access mask. Currently, it will
always get the first one in the list that matches binding/desc_set
regardless of the access mask, but other variables might have different
access mask.

Fix this by being conservative if another variable uses the same
binding/desc_set because we can't get it reliably without adding
a new field to vulkan_resource_index.

This fixes rendering issues in Resident Evil Village with vkd3d-proton.
This bug has been uncovered by ("spirv: Don't remove variables used by
resource indexing intrinsics") because variables are no longer removed

No fossils-db changes.

Cc: 21.1 mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10692>

ci/freedreno: Skip a test that's taking out the a530 boards.

This one has shown up as faulting several times, with CP failures
afterward.

Fixes: e8640fef9089 ("ci: Uprev piglit to 3351e8952 ("max-texture-size: report merged results")")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10848>

drisw: Don't bzero displaytarget pixels

Drivers generally don't clear new resources.

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10758>

gallium/dri: Remove unused dri_drawable::drisw_surface

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10758>

turnip: make cmdstream bo's read-only to GPU

Would allow earlier faults instead of having corrupted cmdstream.

This was already done to Freedreno long ago in:
    04aff7e4 "freedreno: make cmdstream bo's read-only to GPU"

Since private memory should be GPU writable it is now allocated
separately, instead of suballocation from now read-only cmdstream.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10807>

turnip: make possible to create read-only bo with tu_bo_init_new

GPU won't be able to write to such BOs, which would to useful for
cmdstream BOs.

Move "bool dump" to the new flags along the way.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10807>

panfrost: Shorten iffy comment

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10845>

panfrost: Garbage collect comment

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10845>

panfrost: Make clear which limits are arbitrary

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10845>

panfrost: Use natural shader limits

Mostly arbitrary. Use the Gallium limits which are all less than the
hardware limits, and static_assert that this is the case to future
proof.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10845>

panfrost: Drop random #define

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10845>

midgard: Fix type for vertex_builtin_arg() and compute_builtin_arg().

It takes an intrinsic, not an ALU op.  Fixes a clang complaint about enum
conversion.

Fixes: 306800d747bc ("pan/midgard: Lower gl_VertexID/gl_InstanceID to attributes")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10843>

panfrost: Limit the number of active batch to 32

Should improve memory usage at the expense of more agressive batch
submission when the free batch pool is empty. The limit has been chosen
such that a batches bitmap fits nicely in a 32-bit integer.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10842>

panfrost: Simplify the dependency tracking logic

Flush batches when a new batch accessing the same set of BOs comes in
instead of delaying that operation. This greatly simplifies the
dependency tracking logic and shouldn't hurt the perfs (it might even
improve the latency since jobs are now flushed earlier).

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10842>

panfrost: Avoid duplicate entries in access->readers

The PAN_BATCH_DEBUG section was too wide, potentially leading to
double insertion of the same batch in the readers array. Let's get rid
of this already_accessed parameter and make sure
panfrost_batch_update_bo_access() is only called if the RW flags have
changed.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10842>

panfrost: Don't freeze blit batches

We don't want blit batches to interfere with subsequent draw calls, but
we want to keep them in the list of pending batches (so they gets flushed
when glFlush() is called), so calling panfrost_freeze_batch() is not an
option. Reset the ctx->batch pointer manually.

Note that there's no risk to have those batches re-used to queue
unrelated draws because we created a unique pipe_surface object per
blit call, and the batch logic use the surface pointers to calculate
the batch key hash.

Fixes: e1649e916d67 ("panfrost: Use pan_blit() when PAN_MESA_DEBUG=panblit")
Reported-by: Icecream95 <ixn@disroot.org>
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Tested-by: Icecream95 <ixn@disroot.org>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10842>

i915g: fix implicit fallthrough

Closes: #4777
Fixes: 4e861ac4a1f ("i915g: Add more optimizations")
Fixes: f34fd58ec92 ("i915g: implement unfenced relocs for textures using tiling bits")
Fixes: beaf039f972 ("i915g: cleanup static state calculation, part 1")
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10811>

radv: dump the trap handler shader with RADV_DEBUG=metashaders

Help debugging it.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10656>

panfrost: Fix major flaw in BO cache

BOs in the cache are chronological, so we try oldest BOs first. That
means if we find the oldest BO is busy, likely every BO is busy, and we
should bail early. This dramatically reduces the useless cycles spent in
bo_wait.

I studied the BO cache of the following drivers, all of which handle
this correctly: iris, lima, etnaviv, freedreno, vc4, v3d, v3dv.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10794>

ci: Build asahi in meson-gallium job

Adding full CI for Asahi won't make sense for a while, but simply
build-testing should catch a lot of problems with whole tree refactors
in Gallium and NIR.

Closes: #4744
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10826>

Revert "CI: Disable all Panfrost/AMD/Iris automatic jobs"

This reverts commit e1d74c9cfde45ecff1927c2ebf47fc0f42ab0b6a.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10838>

aco/ra: use flags instead of booleans for update_renames()

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10459>

aco/ra: fix get_reg_for_operand() with vector operands

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10459>

aco/ra: fix get_reg_for_operand() when the blocking var is a vector

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10459>

aco/ra: fix get_reg_for_operand() with no free registers

fossil-db (Sienna Cichlid):
Totals from 195 (0.13% of 149839) affected shaders:
CodeSize: 2352160 -> 2356720 (+0.19%); split: -0.00%, +0.20%
Instrs: 431976 -> 433124 (+0.27%); split: -0.00%, +0.27%
Latency: 10174434 -> 10174897 (+0.00%); split: -0.00%, +0.00%
InvThroughput: 4044388 -> 4044425 (+0.00%); split: -0.00%, +0.00%
Copies: 67634 -> 68762 (+1.67%); split: -0.00%, +1.67%

fossil-db (Polaris):
Totals from 186 (0.12% of 151365) affected shaders:
CodeSize: 2272356 -> 2276848 (+0.20%); split: -0.00%, +0.20%
Instrs: 432390 -> 433513 (+0.26%); split: -0.00%, +0.26%
Latency: 13153394 -> 13160194 (+0.05%); split: -0.00%, +0.05%
InvThroughput: 10889509 -> 10889967 (+0.00%); split: -0.00%, +0.00%
SClause: 12745 -> 12747 (+0.02%)
Copies: 74832 -> 75945 (+1.49%); split: -0.01%, +1.50%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10459>

aco/ra: initialize temp_in_scc earlier

We need to know if there's a temporary in SCC before the instruction, not
after.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: 93c8ebfa780 ("aco: Initial commit of independent AMD compiler")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10459>

zink: add even more validation for linear images before creation

nvidia claims format feature support for various flags but then doesn't
actually support them for a 3D image, only 2D, meaning that an extra step
is necessary before creating a linear image

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10823>

zink: populate images with u_blitter if transfer_dst isn't available

this seems incredibly unlikely, but now it's handled

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10823>