pan/bi: Bundle after RA

Flag day change to swap the order of the "scheduler" with the register
allocator. This gives RA much more freedom without significantly
hndering bundling.

It also opens up the door to Adult-level Scheduling which would occur
prior to bundling.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11123>

pan/bi: Fix bi_rewrite_passthrough ordering

The ordering is irrelevant for SSA form input, but is very relevant for
register input.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11123>

pan/bi: Simplify TEXC codegen for sr_count=0

Obscure case.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11123>

pan/bi: Use TEXS_2D for rect textures

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11123>

pan/bi: Pull out bi_count_write_registers

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11123>

vulkan: fix back compat with Android Oreo and below

buffer_handle_t definition was previously inside the deprecated
system/core/include/system/window.h.

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Acked-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11301>

svga: Initialize pipe_shader_state for transform shaders

This fixes crashes for opengl apps. Issue is found in vmware
internal testing

Fixes: f01c0565bb9 ("draw: free the NIR IR.")

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11274>

util/blitter: remove duplicated set_sample_mask calls

it doesn't make sense to have both, so just keep the simpler one

no functional changes because this was redundant

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11294>

swrast: Fix a warning from gcc 11

gcc 11 dixit:

   In function ‘sample_2d_ewa’,
       inlined from ‘sample_lambda_2d_aniso’ at ../src/mesa/swrast/s_texfilter.c:1995:10:
   ../src/mesa/swrast/s_texfilter.c:1729:13: warning: ‘sample_2d_nearest’ reading 16 bytes from a region of size 8 [-Wstringop-overread]
    1729 |             sample_2d_nearest(ctx, samp, img, newCoord, rgba);
         |             ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   ../src/mesa/swrast/s_texfilter.c: In function ‘sample_lambda_2d_aniso’:
   ../src/mesa/swrast/s_texfilter.c:1729:13: note: referencing argument 4 of type ‘const GLfloat *’ {aka ‘const float *’}

Indeed, newCoord is GLfloat[2] but the argument is typed GLfloat[4],
even though only the first two (s and t) are ever read. Fix the array
size in the function signature to reflect the maximum element actually
addressed.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11273>

nouveau: Don't require RTTI and use it only when enabled

The only case RTTI is used in nouveau is type assertion at:
File src/gallium/drivers/nouveau/codegen/nv50_ir.cpp:

    assert(typeid(*i) == typeid(*this));

This assertion is used 'to be on the safe side' only and not mandatory.

In Android we do not have rtti for libLLVM therefore this assertion
has to be skipped.

Signed-off-by: Roman Stratiienko <r.stratiienko@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11069>

v3dv: enable KHR_uniform_buffer_standard_layout

We already support this memory layout. All relevant CTS tests seem to
pass

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11276>

v3dv: enable KHR_incremental_present

All bits should already be provided by wsi/common.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11276>

v3dv: enable KHR_sampler_mirror_clamp_to_edge

This is already implemented

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11276>

v3dv: enable KHR_image_format_list

There's nothing checking for mutable formats, so this needs no changes

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11276>

nir: document that ACCESS_RESTRICT is not set at intrinsics

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/7295>

nir/opt_load_store_vectorize: only require one variable to be restrict

No fossil-db changes.

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/7295>

nir/opt_load_store_vectorize: check for restrict at the variable

SPIR-V -> NIR doesn't set ACCESS_RESTRICT at the intrinsic.

fossil-db (GFX10.3):
Totals from 3 (0.00% of 139391) affected shaders:
CodeSize: 12364 -> 12356 (-0.06%)
Instrs: 2493 -> 2494 (+0.04%); split: -0.04%, +0.08%
Cycles: 15279372 -> 15295756 (+0.11%); split: -0.11%, +0.21%

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/7295>

nir/load_store_vectorizer: fix check_for_robustness() with indirect loads

fossil-db (GFX10.3, robustness2 enabled):
Totals from 13958 (9.54% of 146267) affected shaders:
VGPRs: 609168 -> 624304 (+2.48%); split: -0.05%, +2.53%
CodeSize: 48229504 -> 48488392 (+0.54%); split: -0.02%, +0.56%
MaxWaves: 354426 -> 349448 (-1.40%); split: +0.00%, -1.41%
Instrs: 9332093 -> 9375053 (+0.46%); split: -0.03%, +0.49%

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/7295>

v3dv: document two supported extensions

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11268>

radv: increase maxComputeSharedMemorySize

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11262>

aco: fix emit_mbcnt() with a VGPR mask

Found by inspection. Should be possible with nir_intrinsic_mbcnt_amd.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11295>

v3dv: implement VK_KHR_get_display_properties2

This is entirely implemented in the common WSI code, we just need to
implement the API entry points.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11287>

glapi: fix Warray-parameter

Fixes these new gcc11 warnings:

In file included from ../src/mapi/glapi/glapi_dispatch.c:174:
src/mapi/glapi/gen/glapitemp.h:3191:68: warning: argument 1 of type 'const GLdouble *' {aka 'const double *'} declared as a pointer [-Warray-parameter=]
 3191 | KEYWORD1 void KEYWORD2 NAME(LoadTransposeMatrixd)(const GLdouble * m)
      |                                                   ~~~~~~~~~~~~~~~~~^
In file included from ../src/mapi/glapi/glapi_priv.h:31,
                 from ../src/mapi/glapi/glapi_dispatch.c:40:
../include/GL/gl.h:1901:62: note: previously declared as an array 'const GLdouble[16]' {aka 'const double[16]'}
 1901 | GLAPI void GLAPIENTRY glLoadTransposeMatrixd( const GLdouble m[16] );

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11198>

aco: Use as_vgpr for the second source of mbcnt_amd.

Fixes: 1e49018cedf700fd21fe1498d24742cecf4b5ff4
Signed-off-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/11292>

ci: add expected list of failures for Bonaire (RADV)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11239>

zink: fix caching of shader variants with inlined uniforms

attempting to read the inlined uniforms directly after the variant key
using the size of the variant is not going to work since the variant union
is (sometimes) much larger than the size of the actual struct being used,
meaning that this would just copy a bunch of zeroes instead of the actual
inlined uniforms

Fixes: 7f28775edcc ("zink: implement uniform inlining")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11003>

radv/winsys: add a small comment explaining the CHAIN bit

Without it the hardware launches an IB2 which might hang in some
rare situations.

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/11214>

radv: do not launch an IB2 for secondary cmdbuf with INDIRECT_MULTI on GFX7

It's illegal to emit DRAW_{INDEX}_INDIRECT_MULTI from an IB2 on GFX7.

PAL applies this workaround for indirect dispatches and also on
GFX8-9 but it doesn't seem needed.

This fixes various GPU hangs on Bonaire (GFX7).

Cc: 21.1 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/11214>

panfrost/ci: Add some failures that crept in

For a few days the CI wasn't reporting failures, and these crept in.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11218>

iris/ci: Update the checksums for the pixmark-piano trace

The rendering changed while the CI was down. The rendering still looks
good, only some pixels in the edges have changed, probably due to
blending?

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11218>

ci/lava: Handle proxy download failures

LAVA doesn't consider failure to download a kernel/initramfs as an
infrastructure error, rather just a user error for supplying a broken
URL. We know our URLs aren't broken (because we're perfect), so assume
that failures in download validation are network issues and retry when
we hit them.

LAVA itself has been fixed to retry internally, so we'll get that when
upgrade in a couple of weeks, but gloss over it for now.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11218>

ci/lava: Iterate all job results, not just the first

Each step in a LAVA job returns separate results; a successfully-retired
job can have about 12 entries. Make sure we iterate through all of them
when we're looking for infrastructure errors.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11218>

ci/lava: do not save lava.yaml in the artifacts

We want to avoid leaking private data used in merge process.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11218>

ci/lava: propely report test failure through sys.exit()

We added lava_job_submitter.py to improve our robusteness, but
the final result reporting was not handled correctly by the script.

This change fix it by properly calling sys.exit() on failures.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11218>

v3dv: don't lower vulkan resource index result to scalar

The intrinsic produces a vec2, so let's honor that and avoid the weird
lowering to scalar and later reconstruction to vec2 when we find
load vulkan descriptor intrinsics.

It fixes tests like this (which require that we expose KHR_spirv_1_4):
dEQP-VK.spirv_assembly.instruction.spirv1p4.opptrequal.null_comparisons_ssbo_equal
that otherwise produce bad code that tries to access a vec2 from the result of
that intrinsic, leading to NIR validation errors.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11257>

v3dv: expose KHR_variable_pointers

We only support the VariablePointersStorageBuffer feature for now,
which is the only one that is mandatory, and for which we seem to
be passing all the relevant tests already.

Exposing the optional VariablePointers feature would require that
we support non-constant indexing on UBO/SSBO first.

Relevant CTS tests:
dEQP-VK.*pointer*

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11257>

u_blitter: fix stencil blit fallback for crocus.

crocus needs the sampler view decl to know it's a uint return value.

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11280>

u_blitter: fix fs used when no color emitted

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11280>

freedreno: Remove assert

In multi-context scenarios, one context writing to a resource can race
with a pctx->flush_resource() on another context/thread.  Which means
that by the end of flush_resource() we can have a new write_batch.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11200>

freedreno: Add tid to DBG() msgs

I keep hacking this up locally when debugging TC sort of issues.  Which
is maybe a sign that we should just add this upstream.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11200>

freedreno: Move assert

The batch can be flushed immediately after unlocking, so this assert
needs to move.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11200>

freedreno: Add string-marker debug trace

This is useful in combination with --markers arg to apitrace.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11200>

gallium/u_threaded: Missing driver-thread marking

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11200>

freedreno: Fix fdperf flush

We created and initialized the fence, but forgot to pass it to
fd_submit_flush().

Fixes: aafcd8aacb8 ("freedreno: Re-work fd_submit fence interface")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11200>

freedreno: Fix batch flush race condition

We need to remove the batch cache entry before marking as flushed.

Note that we are already holding the batch lock, to prevent flushing
something that another context is emitting cmdstream to, but there is
a window between batch cache lookup (under screen lock) and acquiring
the batch lock that could previously result in batch cache lookup
returning a flushed batch.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11200>

iris: Move target_to_isl_surf_dim to iris_resource.c

It's only remaining caller lives there.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11235>

iris: Use isl_surf_get_image_surf instead of hand-rolling it

For the 3D RT read case on BDW, we can use the ISL helper.  For the 1D
array case, we don't actually need to flatten; we can just smash it to a
2D array surface because the layout for 1D and 2D is the same on BDW.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11235>

Revert "st/mesa: execute glFlush asynchronously if no image has been imported/exported"

A number of the piglit glx tests use multiple contexts on a single
thread, and previously the flush in MakeCurrent() was enforcing the
ordering between draws on those different contexts.  When that flush
made ASYNC, now there is nothing ordering the draws because we have
two (or more) driver threads for a single frontend thread which is
using nothing more than glxMakeCurrent() to enforce the ordering.

This reverts commit 057a702a3f6a78a8bcd347a74e5a79d70dfc4153.

Fixes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4903
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11278>

util/vbuf: fix buffer overrun in attribute conversions

using the stride to calculate the buffer map size here is not correct, as
the stride is not necessarily equal to the attribute size, it's only the distance
between elements. for the case of overlapping elements (cts does this), the result
is that the attribute conversion will read past the end of the mapped src region

this is usually fine for drivers when they directly map the vertex buffer, as the memory
past the requested region is usually made available, but in the case where the readback
occurs using a staging resource sized exactly to the map region, this overflows and fails
silently, not even triggering a valgrind error because gpu memory lol

instead, add the size of the largest possible element, which will automatically be clamped
and ensure any staging buffers are correctly sized

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10963>

add execmem build option

The execmem option can be set to false to disable the dynamic
dispatch patching that requires mmap(PROT_WRITE | PROT_EXEC),
which is undesirable on some platforms.

Signed-off-by: John Bates <jbates@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10987>

egl/android: refactor to use the legit vndk/window.h header

Using the system/window.h header can potentially cause AHB breakage
because the system header is reserved for platform internal use.

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11206>

egl/android: fix cached buffer slots for EGL Android winsys

Android WSI api contract requires to allocate min_undequeued_buffers + 2
to achieve "triple buffering" effect, which is when the composer backend
acquired the allowed max numbder of buffers, the producer still has 2
buffers to rotate.

ANativeWindow either belongs to SurfaceView which presents directly to
SurfaceFlinger or belongs to other surfaces from the UI framework. For
the former, SurfaceFlinger hardcodes triple buffering for EGL. For the
latter, the surface caps decide the buffer limits or HWUI intercepts and
adjusts the min_undequeued_buffers to hint the EGL implementation to
prepare enough buffer cache slots while HWUI sets the max dequeued
buffer count accordingly.

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11206>

classic/xlib: Fix the build after !9817

This commit mostly exists to be backported to 21.1.x, because the
classic binding to xlib is soon to be retired.

Fixes: 4daef7ffe37 mesa: Remove redundant gl_config::sampleBuffers
Closes: mesa/mesa#4900
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11270>

zink/ci: increase piglit and deqp-runner timeouts

A bunch of tests are currently taking ~60 seconds, dangerously close to
the current 60 seconds timeout. Add some congesion or just bad luck to
a runner, or some other background-task, and we can very easily be
pushed above that.

Since we're rendering on LLVMpipe rather than HW, this isn't as
unreasonable as it might sound at first. So let's copy a trick from
VirGL's CI playbook, and bump the timeout to 180 seconds, three times
of what we currently use.

That might seem a bit excessive, but it's better to make it *very*
unlikely to cause trouble than a little bit unlikely.

Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11272>

ci: Add Piglit gl-1.0-blend-func to everyone's skips

I've seen this timing out on Zink, v3d, radeonsi, and Panfrost
periodically, but more and more lately. Try to minimise the damage by
just universally skipping it until someone comes up with a better idea.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11267>

ci: Run tests inside Crosvm

Allow running tests within Crosvm for testing the virtio-gpu winsys in
Virgl, tested with Piglit only for now.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10862>

ci: Store the credentials in /tmp

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10862>

ci: Move Kernel build tasks into its own file

Allows to reuse the Kernel build tasks everywhere it is needed.

The x86_test-gl container now need a kernel image to use for the crosvm environment,
Reuse this task there.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10862>

ci: Build Crosvm in our container

And also add the required bits to the x86_64 kernel.

syslogd is needed by Crosvm.

iptables is needed to route packets in and out the VM.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10862>

ci/windows: Re-enable Windows build

The GStreamer-provided Windows build machine is back now.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11261>

ci/windows: Artifact Meson build and test logs

These are useful, especially when tests fail. The other jobs do this, so
fall into line.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11261>

mesa: also flush after compute dispatch when debug flag enabled

This makes MESA_DEBUG=flush also flush after compute dispatches.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11205>

ci/radeonsi: Skip flaky glx-swap-copy test

This seems to be failing a bunch lately, so let's just skip it for now.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11265>

zink: export PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS

dynamic state required because pipeline hashing for this would be too pita

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: use dynamic state to apply sample locations during draw

not even going to attempt to figure out how to efficiently hash this

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: add a pipe_context::evaluate_depth_buffer hook

this works by flagging the next barrier to use the current sample locations
so that everything works as expected during decompression

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: update vk sample location info during framebuffer setup

this needs to be stored as a persistent thing on the ctx so it can be
reused without needing to be recalculated

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: add a util function for populating VkSampleLocationsInfoEXT

this will be reused

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: also flag sample_locations_changed if framebuffer samples changes

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: add a pipe_context::set_sample_locations hook

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink; add a pipe_screen::get_sample_pixel_grid hook

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: set VK_IMAGE_CREATE_SAMPLE_LOCATIONS_COMPATIBLE_DEPTH_BIT_EXT on zs rts

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

zink: populate maxSampleLocationGridSize for all available sample sizes on init

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11189>

amd: Add extra source to the mbcnt_amd NIR intrinsic.

The v_mbcnt instructions can take an extra source that they add to
the result. This is not exposed in SPIR-V but we now expose it in NIR.

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/11072>

ac/nir: Refactor and optimize the repacking sequence.

According to feedback, the terminology with "exclusive scan"
and "reduction" is difficult. Change it to use "repack" instead,
which better fits what this sequence is actually used for.

The new sequence stores only 1 byte / wave to LDS, and uses packed
instructions to produce the results. This has lower latency and
fewer instructions than what we previously had.

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/11072>

aco: Keep VGPR destinations for uniform shared loads when beneficial.

When the result of these loads is only used by cross-lane instructions,
it is beneficial to use a VGPR destination. This is because this allows
to put the s_waitcnt further down, which decreases latency.

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/11072>

aco: Implement byte and lane permute intrinsics.

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/11072>

nir: Add AMD-specific byte and lane permute intrinsics.

These map directly to v_perm_b32 and v_permlane_b32.
Unfortunately there is no corresponding NIR opcode or
intrinsics, and it's too tedious to puzzle these things
together from the existing NIR instructions.

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/11072>

aco: Add validation for v_permlane instructions.

Previously there hasn't been any validation for these instructions,
but after shooting myself in the leg with it a few times, I decided
to add the validation now.

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/11072>

aco: Implement nir_op_sad_u8x4.

Fix up the operand size for v_sad instructions, and implement
the new NIR horizontal add. There is no viable way to do this
in SALU, so let's always use a VGPR destination.

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/11072>

nir: Add nir_op_sad_u8x4 which corresponds to AMD's v_sad_u8.

NIR currently doesn't have any intrinsics for a horizontal packed add,
so this one is modeled after AMD's v_sad_u8.

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/11072>

aco: Add note about v_alignbyte in the ISA README.

We tried to use this instruction for a more optimal sequence,
but it turned out that it doesn't exactly work as it was
supposed to. This note is to help others who want to use it.

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/11072>

llvmpipe: restrict optim bug workaround to gcc 10.x

seems fixed in 11.x, see https://gitlab.freedesktop.org/mesa/mesa/-/issues/3906

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11209>

mesa: always expose NV_image_formats and OES_shader_image_atomic

As a result of some previous changes, it is now possible to expose ES
3.1 without having the ARB_shader_image_load_store enable set. However
we still want those other extensions. When there's a driver that's
capable of exposing ES 3.1 but not these extensions, more caps can be
added.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11204>

zink: remove variable length arrays in ntv

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11171>

util/prim_restart: simplify util_draw_vbo_without_prim_restart a bit

by initilizing variables up front and using actual draw structs for the
rewrite, a multidraw can be passed to the driver

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10966>

util/prim_restart: update index bounds before draws in util_draw_vbo_without_prim_restart

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10966>

util/prim_restart: store the total index count when rewriting draws

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10966>

util/prim_restart: store index bounds while rewriting draws

may as well

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10966>

util/prim_restart: pre-trim degenerate primitives during draw rewrite

these will be eliminated by drivers anyway

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10966>

util/prim_restart: assert the index size at the start of the function

this lets it be removed from the macro loop

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10966>

aco: use ds_read_{u8,u16}_d16

This allows partial writes and writes to the upper half of the destination.

fossil-db (Sienna Cichlid):
Totals from 135 (0.09% of 149839) affected shaders:

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/11113>

aco: don't ever widen 8/16-bit sgpr load_shared

Doesn't seem to create incorrect code, but it is suboptimal.

No fossil-db changes.

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/11113>

radv: improve LDS alignment check for load/store vectorization

Previously, this could vectorize two scalar 16-bit loads into a u8vec4
load.

No fossil-db changes.

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/11113>

aco: use v1b/v2b for ds_read_u8/ds_read_u16

The p_extract_vector isn't necessary.

For ds_read_u8 and ds_read_u16, we used a 32-bit regclass, but did't load
32 bits, and used dst_hint for vector loads when we shouldn't have.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4863
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11113>

ci: update list of expected failures for Pitcairn/Oland (RADV)

The robustness2 failures were a mistake because they are actually
not supported (no VK_EXT_scalar_block_layout on GFX6).

The sparse related failures are no longer supported since sparse
is only enabled for Polaris10+.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11243>

aco: fix emitting literal offsets with SMEM on GFX7

When the offset is negative, reg() isn't 255. Fix this by splitting
SGPR and literal emission. While we are at it, adjust a comment
saying that literals are also accepted on GFX6 which is wrong.

Fixes another batch of robustness tests.

Cc: 21.1 mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11247>

ci/zink: Skip flaky GLX test

!11218 definitely didn't change anything about Zink that should make it
read black back from GLX. Assuming it's a race somewhere, just add it to
skips and move on.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11256>

radv: dump SPIR-V instead of using spirv-dis when generating a hang report

Useful when spirv-dis isn't found.

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/11034>

radv: Implement VK_EXT_global_priority_query.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11215>