aco: optimize load_local_invocation_index with single-wave workgroups

fossil-db (Sienna Cichlid):
Totals from 668 (0.52% of 128647) affected shaders:
CodeSize: 2201912 -> 2193336 (-0.39%)
Instrs: 403124 -> 402325 (-0.20%)
Latency: 4510940 -> 4510214 (-0.02%); split: -0.02%, +0.00%
InvThroughput: 681057 -> 679453 (-0.24%); split: -0.24%, +0.00%
VClause: 6470 -> 6467 (-0.05%)
SClause: 12759 -> 12755 (-0.03%)
Copies: 26348 -> 26218 (-0.49%); split: -0.50%, +0.00%
PreSGPRs: 26140 -> 26101 (-0.15%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel-schuermann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13757>

radv: lower load_local_invocation_index with 1D workgroups

For 1D workgroups, we can just load from an input VGPR.

fossil-db (Sienna Cichlid):
Totals from 226 (0.18% of 128647) affected shaders:
CodeSize: 1200476 -> 1195696 (-0.40%); split: -0.49%, +0.09%
Instrs: 223817 -> 223328 (-0.22%); split: -0.29%, +0.07%
Latency: 2552394 -> 2549606 (-0.11%); split: -0.15%, +0.04%
InvThroughput: 533989 -> 532670 (-0.25%); split: -0.27%, +0.02%
VClause: 5191 -> 5188 (-0.06%)
SClause: 7637 -> 7636 (-0.01%)
Copies: 18165 -> 18182 (+0.09%); split: -0.22%, +0.31%
Branches: 10446 -> 10442 (-0.04%)
PreSGPRs: 8049 -> 8041 (-0.10%); split: -0.17%, +0.07%
PreVGPRs: 7785 -> 7767 (-0.23%); split: -0.32%, +0.09%

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel-schuermann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13757>

nir/lower_system_values: replace local_invocation_id components with zero

fossil-db (Sienna Cichlid):
Totals from 360 (0.28% of 128647) affected shaders:
VGPRs: 7912 -> 7272 (-8.09%); split: -8.59%, +0.51%
CodeSize: 542456 -> 544688 (+0.41%); split: -0.32%, +0.73%
MaxWaves: 10866 -> 10952 (+0.79%)
Instrs: 95973 -> 96010 (+0.04%); split: -0.34%, +0.38%
Latency: 4366023 -> 4344664 (-0.49%); split: -0.90%, +0.41%
InvThroughput: 19656659 -> 18297185 (-6.92%); split: -6.92%, +0.00%
VClause: 3242 -> 3116 (-3.89%); split: -4.04%, +0.15%
SClause: 3422 -> 3504 (+2.40%); split: -0.20%, +2.60%
Copies: 8854 -> 9376 (+5.90%); split: -0.89%, +6.79%
Branches: 2329 -> 2326 (-0.13%); split: -0.39%, +0.26%
PreSGPRs: 7620 -> 7841 (+2.90%); split: -0.43%, +3.33%
PreVGPRs: 5765 -> 5504 (-4.53%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel-schuermann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13757>

tu: Expose Vulkan 1.2

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13756>

tu: Add VK_KHR_buffer_device_address stubs

dEQP-VK.api.version_check.entry_points requires us to return a function
pointer, even though the feature is optional in Vulkan 1.2.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13756>

tu: Enable subgroupBroadcastDynamicId

It's a Vulkan 1.2 only feature, but it's trivially supported.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13756>

freedreno/a[345]xx: add R8/RG8 SRGB formats

These enable the GL_EXT_texture_sRGB_R8 / GL_EXT_texture_sRGB_RG8
extensions.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13765>

freedreno: prefer float immediates when float values are involved

Using double immediates can cause a natively-float value to have to get
upgraded to a double unnecessarily. Use float immediates where possible.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13764>

pan/mdg: Remove duplicate compiler option

Noted by clang.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Enable dual texture fusing pass

Everything is in place for it now -- ship it! Our Bifrost cycle model is
coarse, so take the shader-db results with a massive spoonful of salt:

total instructions in shared programs: 107504 -> 107252 (-0.23%)
instructions in affected programs: 39692 -> 39440 (-0.63%)
helped: 191
HURT: 1
helped stats (abs) min: 1.0 max: 20.0 x̄: 1.32 x̃: 1
helped stats (rel) min: 0.11% max: 9.52% x̄: 1.21% x̃: 0.98%
HURT stats (abs)   min: 1.0 max: 1.0 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 2.04% max: 2.04% x̄: 2.04% x̃: 2.04%
95% mean confidence interval for instructions value: -1.60 -1.02
95% mean confidence interval for instructions %-change: -1.37% -1.01%
Instructions are helped.

total tuples in shared programs: 89864 -> 89664 (-0.22%)
tuples in affected programs: 27787 -> 27587 (-0.72%)
helped: 146
HURT: 6
helped stats (abs) min: 1.0 max: 17.0 x̄: 1.54 x̃: 1
helped stats (rel) min: 0.14% max: 15.38% x̄: 1.83% x̃: 1.25%
HURT stats (abs)   min: 1.0 max: 11.0 x̄: 4.17 x̃: 2
HURT stats (rel)   min: 0.54% max: 3.55% x̄: 1.29% x̃: 0.87%
95% mean confidence interval for tuples value: -1.64 -0.99
95% mean confidence interval for tuples %-change: -2.06% -1.36%
Tuples are helped.

total clauses in shared programs: 18253 -> 18044 (-1.15%)
clauses in affected programs: 5127 -> 4918 (-4.08%)
helped: 164
HURT: 1
helped stats (abs) min: 1.0 max: 19.0 x̄: 1.28 x̃: 1
helped stats (rel) min: 0.78% max: 28.57% x̄: 6.73% x̃: 5.88%
HURT stats (abs)   min: 1.0 max: 1.0 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 2.94% max: 2.94% x̄: 2.94% x̃: 2.94%
95% mean confidence interval for clauses value: -1.50 -1.04
95% mean confidence interval for clauses %-change: -7.42% -5.91%
Clauses are helped.

total cycles in shared programs: 8118.54 -> 8103.88 (-0.18%)
cycles in affected programs: 414.96 -> 400.29 (-3.53%)
helped: 43
HURT: 27
helped stats (abs) min: 0.041665999999999315 max: 4.375 x̄: 0.41 x̃: 0
helped stats (rel) min: 0.24% max: 50.00% x̄: 11.49% x̃: 5.26%
HURT stats (abs)   min: 0.041665999999999315 max: 1.1666639999999973 x̄:
0.10 x̃: 0
HURT stats (rel)   min: 0.43% max: 4.71% x̄: 1.42% x̃: 1.28%
95% mean confidence interval for cycles value: -0.35 -0.07
95% mean confidence interval for cycles %-change: -9.50% -3.53%
Cycles are helped.

total arith in shared programs: 3375.67 -> 3376.42 (0.02%)
arith in affected programs: 345.29 -> 346.04 (0.22%)
helped: 24
HURT: 32
helped stats (abs) min: 0.041665999999999315 max: 0.5833329999999997 x̄:
0.09 x̃: 0
helped stats (rel) min: 0.24% max: 14.29% x̄: 2.82% x̃: 1.50%
HURT stats (abs)   min: 0.041665999999999315 max: 1.1666639999999973 x̄:
0.09 x̃: 0
HURT stats (rel)   min: 0.43% max: 4.71% x̄: 1.42% x̃: 1.28%
95% mean confidence interval for arith value: -0.04 0.07
95% mean confidence interval for arith %-change: -1.19% 0.39%
Inconclusive result (value mean confidence interval includes 0).

total texture in shared programs: 1275 -> 1157 (-9.25%)
texture in affected programs: 725.50 -> 607.50 (-16.26%)
helped: 192
HURT: 0
helped stats (abs) min: 0.5 max: 10.0 x̄: 0.61 x̃: 0
helped stats (rel) min: 2.86% max: 50.00% x̄: 25.20% x̃: 25.00%
95% mean confidence interval for texture value: -0.72 -0.51
95% mean confidence interval for texture %-change: -27.12% -23.27%
Texture are helped.

total vary in shared programs: 537.88 -> 536.12 (-0.33%)
vary in affected programs: 2.75 -> 1 (-63.64%)
helped: 1
HURT: 0

total quadwords in shared programs: 79762 -> 79681 (-0.10%)
quadwords in affected programs: 10261 -> 10180 (-0.79%)
helped: 59
HURT: 18
helped stats (abs) min: 1.0 max: 14.0 x̄: 1.88 x̃: 1
helped stats (rel) min: 0.38% max: 8.20% x̄: 1.95% x̃: 1.43%
HURT stats (abs)   min: 1.0 max: 4.0 x̄: 1.67 x̃: 1
HURT stats (rel)   min: 0.46% max: 8.89% x̄: 2.22% x̃: 1.21%
95% mean confidence interval for quadwords value: -1.57 -0.53
95% mean confidence interval for quadwords %-change: -1.59% -0.37%
Quadwords are helped.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Test dual texture fusing

These patterns are quite tricky, so let's make sure we're testing
adequately.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Add dual texture fusing pass

Bifrost supports a special "dual texture" instruction, sampling from two
textures at once at the same coordinate. Each subinstruction is highly
restricted (a subset of TEXS_2D); together, they are represented by TEXC
with a special dual texture operation descriptor. Add an optimization
pass to fuse these instructions.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Fix up dual texturing registers

This must be done after RA. How delightful. Use the GenXML strategy to
just OR the birds.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Add bi_dual_tex_as_u32 helper

Type safe cast, making dual texture descriptors easier to manipulate.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Support dual texture scheduling

Teach the scheduler about dual texturing to avoid an artifical "must not
last" constraint causing suboptimal scheduling like

clause_1:
ds(0) nbb tex ncph dwb(0)
{
    *NOP t0
    +TEXC.skip t1, r0, r1, 0xf1e00144, @r4
    *NOP t0
    +NOP t1
}

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Use BIFROST_TEXTURE_OPERATION_SINGLE enum

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Add bifrost_dual_texture_operation struct

This is the other state of the texture operation descriptor. We must
pack it in the compiler when fusing dual texturing.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Add bifrost_texture_operation_mode enum

Differentiates single/dual texturing.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Add second destination to TEXC

Used to model dual texturing, which writes to separate sets of staging
registers.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Add secondary staging count

Useful for instructions with two independent sets of staging registers
(like dual source blending or dual texturing).

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Make bi_index padding explicit

Avoids reliance on UB.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

pan/bi: Fix typo in helper invocation analysis

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13723>

mesa: add just a tiny bit of debug info to some _mesa_problem calls

I hit these on a4xx with dEQP-GLES2/3 runs. Having more info on them
would make everyone's life easier.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13766>

nv50,nvc0: expose R8/RG8_SRGB formats for texturing

This enables the GL_EXT_texture_sRGB_R8/RG8 extensions.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13769>

turnip: Enable VK_KHR_separate_depth_stencil_layouts

We now start handling depth/stencil layouts separately when
adding implicit subpass dependancies.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13057>

nir/lower_pntc_ytransform: Support PointCoordIsSysval

Pattern match the point coord sysval and support lowering it as well.
This is required to handle flipped framebuffers on Bifrost. However,
what this pass normalizes to is the opposite of the hardware mode we
used on Bifrost before, so we need to swap modes at the same time to
prevent regressions.

Fixes Piglit glsl-fs-pointcoord and glsl-fs-pointcoord_gles2

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13073>

mesa: add missing state to state string computation

This is an internal state, so does not need to be made available in the
string itself (same as the wpos_y_transform). But it needs to be listed
to avoid the mesa_problem.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13073>

v3dv: don't use a global constant for default pipeline dynamic state

Some of these may change across V3D versions, so it is not practical.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13775>

v3dv: account for multisampling when computing subpass granularity

The granularity is defined by the tile size, which is also determined
by multisampling.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13775>

v3d: use V3D_MAX_DRAW_BUFFERS instead of hardcoded constant

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13775>

meson: Update libelf wrap for Windows

Newer libelf update supports 32-bit Windows.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13749>

radv: Match function definitions to declarations

Fixes compiler errors for 32-bit Windows.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13749>

ac: Align ADDR_FASTCALL with addrlib

Fixes linker errors for 32-bit Windows.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13749>

driconf: support META application

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13686>

radeonsi: add radeonsi_force_use_fma32 driconf option

fma32 only round once so has 0.5UP accuracy. mad32 round twice so
has 1UP accuracy. This accuracy difference sometimes make the result
different at the last bit.

Applications like META need more accuracy for display right result.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13686>

ci/bare-metal: switch to common .baremetal-test-arm64

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13751>

ci/bare-metal: add .baremetal-test-arm64

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13751>

ci/etnaviv: no need to force nir anymore

It is the new default.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13751>

ci/etnaviv: armhf: switch to .baremetal-test-armhf

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13751>

ci/bare-metal: armhf: move BM_ROOTFS to generic place

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13751>

aux/primconvert: support pipe_context::draw_vertex_state

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13742>

aux/primconvert: break out primconvert internals into util function

this should (ideally) be no functional changes

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13742>

freedreno/a4xx: fix min/max/bias lod sampler settings

This makes a4xx look more like a3xx for these settings. Most importantly
it adds the workaround for allowing the hw to decide between min and mag
filtering. This fixes a number of dEQP texture filtering tests.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13763>

freedreno/ir3: fix setting the max tf vertex when there are no outputs

Fixes dEQP-GLES3.functional.transform_feedback.* on a4xx.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13760>

freedreno: check batch size after the fallback blitter clear

When force-flushing after every draw, this would otherwise hit a NULL
batch in fd_blitter_clear.

Tested on a4xx.

Suggested-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13761>

anv: fix multiple wait/signal on same binary semaphore

We need to guarantee that when vkQueueSubmit() returns the application
can actually wait on a signaled semaphore/syncobj.

When using a thread to do the submission to i915, this gets a bit
tricky in the following case :

   A syncobj is used both as a wait & signal semaphore and has been
   signaled once already. It contains a fence before entering
   vkQueueSubmit().

   This means we need to reset the syncobj to ensure when we return
   from vkQueueSubmit(), the syncobj contains no stale fence.

   Currently in the Anv, the submission thread is in charge of putting
   the new fence in the syncobj and also picks up the wait fence
   directly from the syncobj. This means we can't reset the syncobj
   from vkQueueSubmit().

The solution to this has been pointed by Bas & Jason :

   In vkQueueSubmit(), clone the wait syncobj fence into a new
   temporary syncobj that will be destroy after submission and use
   this temporary syncobj as a wait fence for i915. This allows us to
   reset the original syncobj in vkQueueSubmit().

   For this to work with wait_before_signal behavior, we also need to
   do a wait-on-materialize on binary semaphores from vkQueueSubmit().
   Otherwise the application thread calling vkQueueSubmit() could race
   the submission thread and pick up the wrong fence when cloing.

v2: Use copy semantic for clone_syncobj_dma_fence() (Jason)
    Do the cloning prior to adding the syncobj to anv_queue_submit so
    that if the cloning fails don't have an invalid syncobj in
    anv_queue_submit (Jason)

v3: Fix another syncobj leak (Jason)

v4: Fix invalid argument order (Lionel)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4945
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11474>

gtest: Fix output of array ASSERT/EXPECT macros

Fixes: 015383d1d78 ("gtest: Add mesa-gtest-extras.h with array ASSERT/EXPECT macros")
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13748>

vulkan/util: Handle depth-only formats in vk_att_ref_stencil_layout

From VUID-VkAttachmentReference2-attachment-04755:
 "If attachment is not VK_ATTACHMENT_UNUSED, and the format of the
  referenced attachment is a depth/stencil format which includes both
  depth and stencil aspects, and layout is
  VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_OPTIMAL or
  VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_OPTIMAL, the pNext chain must include
  a VkAttachmentReferenceStencilLayout structure."

We did not check that there even could be a stencil layout
before fetching it.

Fixes a few tests from:
 dEQP-VK.image.depth_stencil_descriptor.depth_read_only_optimal.*

Fixes: 979ea394e56a418f4138b5f440ec0b641cc5fa0e "vulkan/util: Move
helper functions for depth/stencil images to vk_iamge"

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13740>

mesa/st: Disable NV_copy_depth_to_color on non-doubles-capable HW.

The previous doubles check
(https://gitlab.freedesktop.org/mesa/mesa/-/issues/3459) checked that you
didn't have full doubles emulation turned on, but we also need to check
that you have doubles at all (emulated or not) or non-GL4 drivers will
fail.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13743>

v3d: remove static v3d_start_binning

v3dx(start_binning) is just a call to that method, so let's just use
it directly.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13754>

v3d: remove unused include

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13754>

lima: Fix drawing wide lines

GLES2.0 spec allows parts of wide lines and points to be drawn even if
their center is outside the viewport.
Therefore 0x2000 in PLBU_CMD_PRIMITIVE_SETUP has to be set for points.
This is already our default setting as it seems to have no negative
effect when this bit is always set. Points work as expected but lines
don't. It's hard to RE it, because the affected deqp tests also fail
with the blob.

To respect this behaviour for lines and solve another 2 tests, we need
to do a workaround and temporarily extend the viewport by half of the
line width. The scissor rectangle is still equal with the initial
viewport.

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12971>

ac/rgp: add support for dumping SPM data

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13704>

ac: add initial SPM support

SPM is hardware feature that allows us to dump performance counters
at a sampling interval to a buffer. It is used by RGP to report cache
counters.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13704>

v3d: Update prim_counts when prims generated query in flight without TF

In order to implement GL_PRIMITIVES_GENERATED, v3d allocates a small
resource and adds a command to the job to store the prim counts to it.
However it was only doing this when TF was enabled which meant that if
the query was used with a geometry shader but no TF then the query would
always be zero. This patch makes the driver keep track of how many
PRIMITIVES_GENERATED queries are in flight and then enable writing the
prim count if its more than zero.

Fix dEQP-GLES31.functional.geometry_shading.query.primitives_generated_*

v2: Update CI expectations and references to fixed tests in commit log.
v3: - Add comment that GL_PRIMITIVES_GENERATED query is included because
      OES_geometry_shader, but it is not part of OpenGL ES 3.1. (Iago)
    - Update Fixes to commit introducing geometry shaders. (Iago)

Fixes: a1b7c084 ("v3d: fix primitive queries for geometry shaders")
Signed-off-by: Neil Roberts <nroberts@igalia.com>
Signed-off-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13712>

virgl: Allocate qdws after virgl_init_context to avoid leak.

Fix defect reported by Coverity Scan.

Resource leak (RESOURCE_LEAK)
leaked_storage: Variable qdws going out of scope leaks the storage it
points to.

Fixes: 9a7d6a110e2 ("virgl/drm: explicit context initialization")
Suggested-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13609>

microsoft/spirv_to_dxil: Fix non-Windows build.

../src/microsoft/spirv_to_dxil/dxil_validation.cpp: In function ‘bool validate_dxil(dxil_spirv_object*)’:
../src/microsoft/spirv_to_dxil/dxil_validation.cpp:129:12: error: ‘stderr’ was not declared in this scope
  129 |    fprintf(stderr, "DXIL validation only available in Windows.\n");
      |            ^~~~~~

Fixes: 37c366e2830 ("microsoft/spirv_to_dxil: Add DXIL validation to spirv2dxil")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13736>

freedreno/a6xx: Inline remaining fd6_tex_const_0() call.

Less indirection and fixups for figuring out what's going on.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

freedreno/a6xx: Drop an unused tile_mode arg.

I added this in ebaeddcbb3e9 ("freedreno/a6xx: Rewrite the format table
format/swap helpers.")  but it had already become unused through some
bugfixing.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

freedreno/a6xx: Clean up sysmem fb read patching using fd6_view.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

freedreno/a6xx: Use fd6_view for non-buffer image descriptors, too.

This deletes a whole lot of code, but there's a modest drawoverhead perf
loss:

drawoverhead 1-image change -6.48856% +/- 4.28269% (n=50)
drawoverhead 8-image change -5.29195% +/- 2.62549% (n=90)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

freedreno/a6xx: Switch to relying on fd6_view for our texture descriptors.

Having checked the deltas between fdl6_view and what we did before, switch
over to fdl6_view now.

No statistically significant difference on no-hw drawoverhead 8-texture
change (n=50) with the texture cache disabled from this and the previous
commit.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

freedreno/a6xx: Create a fd6_view at sampler view update time.

The goal is to share the same code as turnip for descriptor setup. This
just calls it and cross-checks.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

freedreno: Set layer_first on (2D) resource imports.

Prevents getting a weird layer stride if you ask for it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

freedreno/fdl6: Skip redundant setting of TILE_ALL for NV12.

We already respect the tile_all flag above, and it should be set in tu.
Fixes a mismatch between fdl6_view_init() and gallium.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

util/format: Add G8_B8R8_420_UNORM to match Vulkan.

turnip was playing fast and loose with the name, using the R8_G8B8 format
name but actually setting the descriptors up to read G8_B8R8 like Vulkan
(sensibly) wants.  This caused trouble when trying to make freedreno and
turnip share code.  By having both orderings as format names, we can share
the descriptor code and also confuse readers less.

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

util: Rename PIPE_FORMAT_G8_B8_R8_420_UNORM.

The only user, turnip, was actually treating it as this layout, matching
vulkan's specification of how the planes map to RGB values.  (Y=G means
that Cb=B and Cr=R).

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13443>

util/ra: Fix deserialization of register sets

Set ra_class::regset and ra_class::index  when deserializing.

Fixes: 95d41a3525d ("ra: Use struct ra_class in the public API.")
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13728>

util/ra: Add simple test for register set serialization

Marked as DISABLED since it currently fails.

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13728>

docs: update calendar for 21.3.0-rc5

Add another release candidate as we're not ready for 21.3.0 final just yet.

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13744>

ci/bare-metal: Close serial and join serial threads before exit.

This should fix the intermittent (~1/week) cheza failure where python
complains that a thread tried to do stdio while the main thread has
exited.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13462>

ci/etnaviv: Add more texturing flakes.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13462>

ci/etnaviv: Mark the rest of uniform_api.random as flaky.

Another 3 new flakes happened on my last run.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13462>

mesa/main: Fix use of alloca() without #include "c99_alloca.h"

Fixes: c216f193 ("mesa: use alloca in search_resource_hash")
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13735>

freedreno: Fix constant-index assumptions in IBO loads.

The encoder already sets up our IBO accesses as potentially nonuniform, so
we just need to be careful to not try to force the IBO index into an
immediate.

Fixes assertion failures in piglit arb_shader_image_load_store-invalid
(intermittent due to
https://gitlab.freedesktop.org/mesa/piglit/-/merge_requests/597), which
had some interesting actual failures hidden behind it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13601>

freedreno: Fix the uniform/nonuniform handling for cat5 bindful modes.

We can see from the dynamically_uniform (compiler doesn't know if you're
uniform or not) vs uniform (compiler can see it's uniform) case in the
blob which is which.  Now that we have the right names, also use the
nonunif flag for encoding the actual non-uniform mode (previously, we were
always setting it always in a way that meant uniform).

I verified this behavior back to a418 with samplers.  The a3xx blob I have
only does GLES3, so we don't have the opaque_type_indexing tests to see.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13601>

anv: don't forget to add scratch buffer to BO list

We reference the scratch BO using a bindless index in the command
streamer instructions, but we forgot to add them to the BO list.

v2: Make use of pipeline reloc list (Jason)

v3: Don't add NULL BOs to the reloc list (Lionel)

v4: Don't add BOs twice to reloc list when dealing with addresses
    (Lionel)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: eeeea5cb873b40 ("anv: Add support for scratch on XeHP")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13544>

aco: Work around MSVC restrict in c99_compat.h

Future LLVM header leads to __declspec(__restrict), which is invalid.
Just undefine the restrict macro to keep __declspec(restrict).

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13505>

wsi/x11: support depth 30 visuals

There's only really one format that makes sense here, since there's no
sRGB equivalent for 10-bit packed formats.

It's possible that this results in flipped colors, if 10-bit visuals
exist that have the channels in the opposite order. (They don't seem to,
on my end)

Perhaps a more principled solution would also compare the exact r/g/b
bitmasks. But for now, this works.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3537
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9450>

vulkan/wsi: Unlock before return on error path.

Fix defect reported by Coverity Scan.

Missing unlock (LOCK)
missing_unlock: Returning without unlocking wsi->wait_mutex.

Fixes: 4885e63a6d2 ("vulkan/wsi: implement missing wsi_register_device_event")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13698>

v3dv,v3d: don't store swizzle pointer in shader/pipeline keys

We had been storing pointers to a driver owned swizzle table
rather than storing the actual swizzle value in various shader
and pipeline keys on both GL and Vulkan drivers.

This doesn't look very robust, particularly since we also
compute sha1 hashes from these values and we may store these
hashes to disk (for the disk cache).

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13738>

vulkan, radv: Support backslash in ICD paths

Vulkan loader wants backslash for paths on Windows. Need to jump through
hoops because Meson does not support backslashes in commands.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13461>

radv/sqtt: fix GPU hangs when capturing from the compute queue

S_008D20_FINISH_DONE is a mask of queues and 1 means "wait on the gfx
queue until the value is not 0" which can never happen when the driver
captures from compute. Instead, use the full mask of possible queues.

Cc: 21.3 mesa-stable
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/13694>

zink: update gfx pipeline shader module pointer even if the program is unchanged

this is used for pipeline comparisons, so it has to always be accurate

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13727>

zink: be more consistent about applying module hash for gfx pipeline

this was a little spaghetti-ish: the module hash was sometimes being applied
during module update, sometimes in draw during program create, and then also
it was removed when a shader unbind would cause the program to no longer be reachable

now things are more consistent:
* keep removing module hash when program becomes unreachable
* only apply module hash in draw during updates there

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13727>

zink: ci updates

these don't spend forever in llvmpipe optimization passes anymore

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13727>

zink: always inline uniforms when running on a cpu driver

the overhead from creating new inlined shader variants is likely to be less than
the time required to fully optimize and run those variants, so just
inline 100% of the time to cut down shader runs

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13727>

zink: implement cs uniform inlining

this implements shader variants for compute

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13727>

zink: radv ci updates for 1dshadow stuff

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13583>

zink: add 1DShadow sampler handling for drivers (radv) that don't support it

some drivers won't create zs textures in any shape but 2D. this can be
handled instead by using 2D textures and then performing shader rewrites to
convert shadow samplers for 1D and 1DArray types to 2D/array

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13583>

util/xmlconfig: Allow DT_UNKNOWN files

Some filesystems don't fill in d_type in dirent. Resulting in
DT_UNKNOWN. Pass this entry through to the next step and use stat on the
full filepath to determine if it is a file.

sshfs is known to not fill d_type.
This resolves an issue where driconf living on an sshfs path wasn't
working.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13697>

anv: Also disallow CCS_E for multi-LOD images

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4616
Fixes: e3101c96bb64 ("anv/image: Disable multi-layer CCS_E on TGL+")
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13680>

zink: add another compiler pass to convert 64bit vertex attribs

gallium always provides uint types, so rewrite the shader to load a 64bit
attrib and then cast back to whatever it was before

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13566>

zink: simplify 64bit vertex attrib lowering

this was a cool myfirstcompilerpass.exe but there's easier ways to do
things like this

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13566>

zink: declare int/float size caps inline with type usage

this is much more accurate than trying to use shader info

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13566>

ci: Add RADV to Android CI.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13164>

util: Add support for clang::fallthrough.

Looks like the __attribute__ version doesn't work for C++ in the
Android build. Only found now because we don't enable
-Wimplicit-fallthrough by default project wide for C++. Only
ACO enables it.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13164>

amd/addrlib: Ignore self-assign warnings.

There was a preference to not fix addrlib to make syncing with
upstream easier.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13164>

meson: Check arguments before adding.

-static-libstdc++ doesn't exist on the Android NDK, casuing all
later has_argument calls to return false even though the compiler
supports that argument.

Fixes: 3aee462781a "meson: add windows compiler checks and libraries"
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13164>

aco: Remove useless sub-expr.

../src/amd/compiler/aco_instruction_selection.cpp:11915:83: error: expression result unused [-Werror,-Wunused-value]
      bld.vop2(aco_opcode::v_lshrrev_b32, fetch_index_def, div_info, instance_id).instr;

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13164>