venus: refactor vn_queue_submission_count_semaphores

Add vn_queue_submission_count_batch_semaphores that works on a single
batch at a time.  Also avoid code duplication.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10290>

gallium: Define PIPE_ARCH_AARCH64 for MSVC arm64 builds

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10014>

util/format: Add some NEON intrinsics-based u_format_unpack.

In looking at the profile of dEQP, GLES3 was spending 5-10% of its time in
ReadPixels, and almost all of that is b8g8r8a8_unorm8.  It's really slow
because we're getting about 47MB/s by doing uncached reads 32 bits at a
time in the code-generated unpack.  If we use NEON to generate larger bus
transactions, we can speed things up to 136MB/s.  In comparison, raw
ldr/str read/writes with no byte swapping can hit a max of 216MB/sec.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10014>

util: Switch the non-block formats to unpacking rgba rows instead of rects.

We have only a few callers of unpack that do rects, so add a helper that
iterates over y adding the strides.  This saves us 36kb of generated code
and means that adding cpu-specific variants for RGBA format unpack will be
much simpler.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10014>

mesa: Remove dead _mesa_unpack_rgba_block().

Despite many changes over its lifetime, it appears to have been dead ever
since its introduction in 2011.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10014>

radeonsi: Report multi-plane formats as unsupported

https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6693 added
NV12 lowering to PIPE_FORMAT_R8_G8B8_420_UNORM, which regressed some
video decode use cases in radeonsi, for example CtsDecodeTestCases in
android CTS. There are also discolored frames in video playback use
cases (i.e. youtube).

https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3632
previously fixed similar issues in radeonsi. This change takes that a
step further to report any multi-plane format as unsupported.

Fixes: 826a10255f5 ("st/mesa: Add NV12 lowering to PIPE_FORMAT_R8_G8B8_420_UNORM")
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Simon Ser <contact@emersion.fr>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9490>

turnip: implement VK_KHR_shader_terminate_invocation

OpTerminateInvocation provides the behavior required by the GLSL
discard statement, which we already implement.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9460>

turnip: implement VK_EXT_shader_demote_to_helper_invocation

The "demote" intrinsic has the semantics of D3D discard, which means
it doesn't change the control flow, allowing derivatives to work.

On A6xx there is no known way to check whether invocation was demoted,
thus we use nir_lower_is_helper_invocation.

Add "logical" OPC_DEMOTE which is later translated to "kill".
Such separation is necessary to run "kill" specific optimizations
which are invalid for "demote".

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9460>

nir: add lowering pass for helperInvocationEXT()

Some hardware doesn't have a way to check if invocation was demoted,
in such case we have to track it ourselves.
OpIsHelperInvocationEXT is specified as:

 "An invocation is currently a helper invocation if it was originally
  invoked as a helper invocation or if it has been demoted to a helper
  invocation by OpDemoteToHelperInvocationEXT."

Therefore we:
- Set gl_IsHelperInvocationEXT = gl_HelperInvocation
- Add "gl_IsHelperInvocationEXT = true" right before each demote
- Add "gl_IsHelperInvocationEXT = gl_IsHelperInvocationEXT || condition"
  right before each demote_if

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9460>

ci/lava: Point the shader cache at tmpfs.

We really don't want to be going out over NFS and back for our shader
cache references.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10297>

ci/piglit: Always include the HTML summary in a run.

It's the best visualization for run results when using the piglit
framework, and now all the DUTs return artifacts.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10297>

ci/lava: Return the run's results/ artifacts from the DUTs.

Finally LAVA users will be able to see deqp XMLs on failures from the
job's artifacts browser.  This replaces a couple of one-off minio uploads
in the piglit runner.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10297>

ci/lavapipe: Don't include deqp's shader_cache in the artifacts.

We don't care to look at it and we don't reuse it.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10297>

aco: fix opquantize2f16 on GFX6-7

Make sure to preserve signed zeroes.

Fixes dEQP-VK.spirv_assembly.instruction.compute.opquantize.flush_to_zero
on GFX6 (Pitcairn). Untested on GFX7.

Fixes: 54a09545ec6 ("aco: optimize a*0.0")
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/10319>

ir3: Assemble and disassemble swz/gat/sct

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10291>

ir3: Improve cat1 modifier disassembly

Remove bit that shouldn't be part of (rptN), and rewrite the handling of
(even) and (pos_infinity) to uncover a missing (neg_infinity) modifier.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10291>

ir3/parser: Fix oob write with immediates array

immediates_count and immediates_size are supposed to have the same
units, but it was only incrementing immediates_count by 1. While we're
here, also fix the case where constants are specified out-of-order.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10291>

microsoft/clc: If local size isn't specified either in the shader or at runtime, set it to (1,1,1)

Otherwise we can end up in situations like having divide-by-zero. If the optimization is smart enough
that we end up with a *constant* divide-by-zero, then the DXIL validator will fail to sign, which
can trigger fatal errors with CLOn12.

We want to run an initial translation of all kernels during program build, but at that point we don't
know the local size to be able to specify it through kernel specialization data.

v2: Metadata output of 0 is used to indicate that the size wasn't explicitly specific. Copy the
    size to the metadata before overriding it to (1,1,1). If conf was explicitly specified,
    update the metadata again (though nobody should be paying attention to it).

Closes: https://github.com/microsoft/OpenCLOn12/issues/20
Closes: https://github.com/darktable-org/darktable/issues/8700
Reviewed-By: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10303>

virgl: resources without any binding can be cached

Certain games create and destroy lots of resources without binding them.
This can take quite a bit of time and even create unneeded
synchronization. However, we know that if a resource was never bound to
anything, it can be cached. This change does that.

Counting the number of uncached allocation with a tabletop simulator trace:
Before: 2967 uncached allocations over the replay
After: 24 uncached allocations over the replay

Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10225>

nir/lower_tex: do not stumble on 16-bit inputs

If a has been lowered to float16 here, then we end up trying to
construct a vector of mixed precision, which the validator asserts
about.

So let's make sure we use the same type for all arguments.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10201>

lavapipe: implement VK_EXT_custom_border_color

This fixes a bunch of piglit failures on Zink.

Reviewed-by: Joshua Ashton <joshua@froggi.es>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10320>

lavapipe: resolve border-color when creating sampler

This is going to make it easier to implement the custom border-color
extension.

While we're moving the code, tweak the memset code a bit, so we don't do
any float-ism in the int-case. It doesn't change anything functionally,
just makes it slightly clearer what's going on here.

Reviewed-by: Joshua Ashton <joshua@froggi.es>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10320>

st/pbo: use cso_set_vertex_buffers_and_elements() for st_pbo_draw

this lets us use the more consistent codepath and ordering (buffers -> elems)
as well as set take_ownership=true to skip the unref

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10257>

zink: fix tcs input reservation for user vars

tcs user vars are var_size[32], which isn't actually how many slots they need,
just how big the variable is (oops), so this needs to be divided
by MAX_PATCH_VERTICES to get the real slot count

slot mapping has always been broken for all tcs inputs, but this probably fixes
all of the related issues there, including unlimited crashes when playing Tomb Raider

Fixes: 2d98efd3232 ("zink: pre-populate locations in variables")

Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10269>

zink: fix tcs slot map eval for user vars

tcs user inputs need to have their size adjusted in order to determine whether
they'll overflow the existing slot map

Fixes: 5c5e1abea24 ("zink: evaluate existing slot map during program init and force new map as needed")

Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10269>

zink: always copy the nir shader before compiling

nir_convert_from_ssa and assign_io_locations both modify this unconditionally,
the latter of which possibly re-modifies variables in ways that can break the
slot map and cause stack overflows during vk driver pipeline compilation

Fixes: 2b4609b66cd ("zink: run nir_convert_from_ssa last during compile")
Fixes: 2d98efd3232 ("zink: pre-populate locations in variables")

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10269>

zink: fix stencil-export cap emission

This cap should only ever be emitted for fragment-shaders, but we
accidentally emit it for all shaders. Let's tighten the check to avoid a
validator warning when emitting non-fragment shaders without support for
VK_EXT_shader_stencil_export.

Fixes: 8724d4fb361 ("zink: check shader stencil output")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10249>

gallium/util: add easy profiling helpers using TIME_ELAPSED queries

There are no users and that's intentional.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10262>

Revert "r600: don't set an index_bias for indirect draw calls"

This reverts commit acdf1a123436cb3fa7d4bbc9cfbb9566c7693de4.

While this commit fixed the gles CTS regressions, it introduced
regressions that made the driver unusable, hence the revert.

Closes #4657

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10311>

vulkan/util: cast enums to int64_t in switch

With the new Vulkan Video extensions a bunch of enum values have been
added. The problem is that those are behind #ifdef so our generated
code complains that we using enum values not defined.

Since we're not using enum names but integer values, we can silence
all those by casting the enum type to int64_t.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10304>

iris: move target to isl dim translate to inline.

This avoids a conflict in namespace when crocus wants to use the
same function.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10308>

iris: move get_time into a static in bufmgr code.

This ends up polluting the namespace if you build iris/crocus
at once, just move it to where it's used for now.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10308>

freedreno/fdl: Re-indent

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/perfcntrs: Re-indent

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/ir2: Re-indent

clang-format -fallback-style=none --style=file -i src/freedreno/ir2/*.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/drm-shim: Re-indent

clang-format -fallback-style=none --style=file -i src/freedreno/drm-shim/*.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/decode: Re-indent

clang-format -fallback-style=none --style=file -i src/freedreno/decode/*.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/computerator: Re-indent

clang-format -fallback-style=none --style=file -i src/freedreno/computerator/*.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/common: Re-indent

clang-format -fallback-style=none --style=file -i src/freedreno/common/*.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/afuc: Re-indent

clang-format -fallback-style=none --style=file -i src/freedreno/afuc/*.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

freedreno/drm: Re-indent

clang-format -fallback-style=none --style=file -i src/freedreno/drm/*.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10293>

radeonsi: fix automatic DCC retiling after compute image stores

Only internal compute shaders use DCC stores, so the TODOs are not
critical yet.

Fixes: 1d64a1045ea - radeonsi: enable dcc image stores on gfx10+

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10261>

radeonsi: fix automatic DCC retiling after DCC clear and DCC decompression

Fixes: d4f7962d48b - radeonsi: Add displayable DCC flushing without explicit flushes.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10261>

radeonsi: fix incorrect comments in culling code and NIR lowering

The lowering code removes the "VS inputs" item from the list because the hw
doesn't support indirect indexing of VS inputs.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10261>

radeonsi: don't decompress DCC for float formats in si_compute_copy_image

Format reinterpretation disabled it.

Fixes: 1d64a1045ea205ee0297d2f741a824811570fc6d "radeonsi: enable dcc image stores on gfx10+"

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10261>

amd/registers: rename IMG_FORMAT to GFX10_FORMAT to disambiguate the meaning

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10261>

amd/registers: clean up gfx103.json

because gfx103.json is automatically generated and can't be changed
manually. This fixes the file generator without changing the generated
header.

Missing registers must be in registers-manually-defined.json, and
missing fields must be in parse_kernel_headers.py.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10261>

amd/registers: fix the kernel header parser with latest headers

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10261>

venus: add vn_wsi_create_scanout_image

A helper called from vn_wsi_create_scanout_image to hijack wsi scanout
image creation.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10294>

venus: add vn_image_create

A helper called from vn_CreateImage to allocate and initialize vn_image.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10294>

venus: add vn_image_init_memory_requirements

A helper called from vn_CreateImage to initialize memory requirements.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10294>

nir: Generate load_ubo_vec4 directly for !PIPE_CAP_NATIVE_INTEGERS

The prog_to_nir->NIR-to-TGSI change ended up causing regressions on r300,
and svga against r300-class hardware, because nir_lower_uniforms_to_ubo()
introduced shifts that nir_lower_ubo_vec4() tried to reverse, but that NIR
couldn't prove are no-ops (since shifting up and back down may drop bits),
and the hardware can't do the integer ops.

Instead, make it so that nir_lower_uniforms_to_ubo can generate
nir_intrinsic_load_ubo_vec4 directly for !INTEGER hardware.

Fixes: cf3fc79cd0ab ("st/mesa: Replace mesa_to_tgsi() with prog_to_nir() and nir_to_tgsi().")
Closes: #4602
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10194>

nir_to_tgsi: Use ARL instead of UARL in the !native_integers case.

Our "integer" index is stored as a float in this case, and we just need to
use teh right opcode for loading it, which will be the only one supported
by !native_integers hardware.

Fixes: cf3fc79cd0ab ("st/mesa: Replace mesa_to_tgsi() with prog_to_nir() and nir_to_tgsi().")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10194>

zink: compare against screen batch id when determining which semaphore to use

semaphores are per-screen now, so check the screen value

Fixes: fa36a16c684 ("zink: make timeline semaphores per-screen")

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10267>

iris/gen12: Work around push constant corruption on context switch.

This has been confirmed to fix sporadic graphics corruption on Gen12
platforms for a number of workloads (including Heaven, Valley and
CS:GO among others).  Corruption seems to occur during context switch
fairly consistently, but unfortunately this problem doesn't seem to be
documented.  Until the hardware team comes up with a better
workaround, fix the problem by reemitting constants at the beginning
of each batch.

No corruption has been observed so far in GL due to preemption,
however this is a possibility to keep in mind, it may be necessary to
disable preemption in addition to this patch in order to fully address
this problem (see also 81201e461772a06ed2).

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4412
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4454
Cc: mesa-stable
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

microsoft/compiler: Propagate access when lowering SSBO loads

We'll want to preserve whether the load is accessing a read-only
SSBO or not.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Support raw SRVs in addition to typed SRVs

These will be used for read-only SSBOs

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Store nir_shader in the ntd_context

This means we can stop passing it as a second parameter all the time

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Better support UBO/SSBO references to descriptors

Previously UBOs only supported static indices, and SSBOs only
supported dynamic indices. UBO support for descriptors was added
as an alternative to static indices, but the logic for detecting
descriptors to SSBOs couldn't just differentiate on constants vs not.

Add a helper which can differentiate cleanly across the board and
handle pre-created handles from descriptors, or static/dynamic raw
indices.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Split Vulkan resource_index / descriptor processing

Instead of doing all of the handle logic in the descriptor load, split
it so that the resource index is actually computed during resource_index
processing, and it's converted to a handle during the load_descriptor.

At the same time, add SSBO handling and dynamic indexing handling.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Emit SSBO variables

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Emit CBVs via variables for Vulkan

The resources need to be emitted in a particular order, so CBVs
have to be emitted first and can't be emitted as we iterate through
instructions.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Support arrays of UBOs

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/compiler: Support raw SRVs/UAVs through dxil_module_get_res_type

Switch the OpenCL "globals" array to use that helper

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/spirv_to_dxil: Lower loads/stores to DXIL

This adjusts UBO loads to be float4 loads, and handles SSBO accesses
that are larger/smaller than 32bit.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

microsoft/spirv_to_dxil: Lower samplers from deref to index

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10149>

egl/android: include "util/compiler.h" for FALLTHROUGH macro

Fixes the following building error:

external/mesa/src/egl/drivers/dri2/platform_android.c:1263:10: error: use of undeclared identifier 'FALLTHROUGH'
         FALLTHROUGH; /* for pbuffers */
         ^
1 error generated.

Fixes: 2928c21eb73b ("Convert most remaining free-form fall-through comments to FALLTHROUGH")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10264>

st/nine: Fix compilation error on non-x86 platforms

The unused variable was improperly declared.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4647
Fixes: d9e79bfe4f ("st/nine: Disable fpu exceptions during init")

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10263>

ci: Add test which occasionally times out to lavapipe-vk skips

E.g. https://gitlab.freedesktop.org/mesa/mesa/-/jobs/8912925 .

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10279>

frontends/va: improve surface attribs processing

Instead of checking whether the attribute is settable for each
attrib type, check that once at the beginning of the loop.

Instead of having an if for each attrib type, use a switch.

Return an error if we encounter an unknown attribute. This allows
the caller to make sure settable attributes aren't ignored. The
intel media driver seems to just assert [1] that it doesn't encounter
unknown attributes.

[1]: https://github.com/intel/media-driver/blob/95d413e519a980f6a7178880ccbfedace092316c/media_driver/linux/common/ddi/media_libva.cpp#L2530

Signed-off-by: Simon Ser <contact@emersion.fr>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10104>

v3dv: better tracking of dirty push constant state

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10283>

v3dv: dirty viewport doesn't affect fragment shaders

The uniform state for the viewport is only used with geometry stages.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10283>

v3dv: improve dirty descriptor set state tracking

We were using the pipeline layout to discard uniform updates for
stages that don't use descriptors, but we can do better by keeping
track of the stages used by the specific dirty descriptor sets and
only update uniforms for stages that are included in those.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10283>

anv: disable baked in pipeline bits from dynamic emission path

In 27ee40f4c9d86e ("anv: Add support for sample locations") we
introduced the ability to emit sample locations baked in as part of
the pipeline or dynamically.

This is different from the previous dynamic states that were always
removed from the pipeline batch and instead emitted dynamically all
the time.

The mistake in 27ee40f4c9d86e is that sample locations are now emitted
all the time, leading to bigger command buffers for unnecessary
reasons.

This change introduces a bit fields of what is baked in the pipeline
and doesn't need to be dynamically emitted.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 4ad4cd89069bfc ("anv: Enabled the VK_EXT_sample_locations extension")
Cc: <mesa-stable>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10282>

anv: fix 3DSTATE_MULTISAMPLE emission on gen8+

When pipeline->dynamic_state.sample_locations.samples is not set
because the state is dynamic, we're currently calling
genX(emit_multisample) with a 0 samples value which is incorrect.

Found when using renderdoc with the drawing overlay.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 4ad4cd89069bfc ("anv: Enabled the VK_EXT_sample_locations extension")
Cc: <mesa-stable>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10282>

v3dv: avoid dereferencing null value

Fixes CID#1468079 "Dereference null return value (NULL_RETURNS)"

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10280>

radv: fix clearing DCC-compressed e5b9g9r9 images

Fixes
dEQP-VK.api.image_clearing.core.clear_color_image.2d.optimal.single_layer.e5b9g9r9_ufloat_pack32_33x128
with RADV_DEBUG=forcecompress on GFX10.3.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: 21.1 <mesa-stable>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10176>

glx: Remove some ancient backwards-compatibility typedefs

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Fold client_gl_only array into its one real user

The initialization we're doing for it in __glXExtensionsCtr is trivial,
and this is only to make glGetString(GL_EXTENSIONS) work in indirect
contexts anyway.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Enable pure-client-library extensions explicitly

These are rare enough to not be worth tracking separately. Especially
since after this change all the known_gl_extensions have N for both
direct_support and direct_only (unsurprising, since that's only used to
compute usable indirect extensions).

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Remove redundant client_support field from extension table

It's always true, simplify accordingly.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Generalize __glXGetStringFromTable a little

Make the second argument optional, and treat its absence as if the
filter always matches.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Remove some non-functional GL extension from the table

We don't actually have GLX protocol for these in Mesa, hence the initial
N column.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Mark GLX_{ATI_pixel_format_float,NV_float_buffer} as supported

... at least as far as libGL is concerned, because they'd work just fine
if you tried. The only thing the ATI extension seems to add (I can't
find an official spec, this is inferred from the registry XML) is
selecting the GLX_RENDER_TYPE, which we don't validate before putting on
the wire. The only thing the NV extension adds is an additional fbconfig
attribute, and that only known by glXGetFBConfigAttrib; our
implementation of that just reads the value the server sends, if any,
and doesn't try to filter out unknown attributes.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Remove major/minor version tracking from extension table

With the previous ancient dead code removed, there's no longer anything
using these values.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Remove some truly ancient unused code

The "left here but disabled" comment dates to 2004! The idea here is to
add extensions implied by a particular GL version to the GL extension
string, but nothing useful is accomplished by doing so, and this is all
only used in the case of indirect rendering anyway.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

glx: Remove silly __glXGetGLVersion() indirection

Hardcode the values at the one caller that uses them.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10223>

radv: keep DCC compressed for clears on compute with image stores

Without image stores, DCC is always decompressed on compute.

Cc: 21.1 <mesa-stable@lists.freedesktop.org>
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/10168>

amd: remove some references to older LLVM versions in comments

Reviewed-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/10199>

amd: drop support for LLVM 10

It doesn't support RDNA 2.

Reviewed-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/10199>

amd: drop support for LLVM 9

This would be easy to support except that it doesn't support RDNA 2.

Reviewed-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/10199>

amd: drop support for LLVM 8

It doesn't support Navi1x and the removal enables this nice code cleanup.

v2: rebase - mareko

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com> (v1)
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-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/10199>

ci: don't build clover with LLVM 9 on radeonsi because it's unsupported

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10199>

intel/eu: Add instruction compaction support on XeHP.

This patch includes a number of reworks and fixes squashed in by
Nanley Chery, Sagar Ghuge, Jordan Justen and Francisco Jerez.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10000>

intel/fs: Add more efficient fragment coordinate calculation.

The PIXEL_X/Y opcodes used by the current implementation are broken on
XeHP due to the new regioning restrictions of the floating-point pipe.
We could have the regioning lowering pass fix it in theory by lowering
the conversions into separate MOV instructions, but that would be more
costly than this implementation that only needs a pair of pipelined
ADDs and a pair of pipelined MOVs.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10000>

intel/fs: Introduce lowering pass to implement derivatives in terms of quad swizzles.

Unfortunately the funky Align1 regions used by the code generator in
order to implement derivatives efficiently aren't available to the
floating-point pipeline on XeHP.  We need to lower them into a number
of pipelined integer shuffle instructions followed by the
floating-point difference computation.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10000>

intel/compiler: Lower txd for 3D samplers on XeHP.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10000>

intel/fs: End computer shader with message gateway on XeHP.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10000>

intel/fs: Disable 3-src immediates on XeHP.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
[ Francisco Jerez: Add TODO comment explaining why this is helpful and
  how we could better fix it. ]

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10000>

intel/eu: Allow 64-bit registers on XeHP.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10000>