llvmpipe: consolidate legacy_points and multisample state

This just makes the code a bit easier to follow.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17653>

llvmpipe: merge all rasterizer-bind functions

This makes it more obivous when what state changes, and they are always
just called in order.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17653>

lavapipe: accurately set image/ssbo access based on shader usage

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15286>

crocus: fix leak in query code.

Fixes: f3630548f1da ("crocus: initial gallium driver for Intel gfx 4-7")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17731>

lavapipe: EXT_non_seamless_cube_map

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17711>

zink: avoid crashing when bindless texture updates aren't used

some apps trigger the texture update path far in advance of when the
texture will be used, so don't crash and wait to do the update

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17665>

zink: assert !rp_changed after renderpass has started

avoid weirdness

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17665>

zink: only flag fbfetch as rp update if fbfetch wasn't already configured

otherwise this is probably just a swapchain acquire which needs to update
the descriptor during begin_renderpass

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17665>

r600/dri-tweak: Disable ARB_shader_image_load_store for "Tomb Raider 2013"

When "Depth of Field" is set to Ultra, a compute shader is emitted that
results in Hardware hangs when OpenGL > 4.3 is available.

If the option is enabled, the game will hang at the menu screen so that
it is no longer possible to simply change the option back. To avoid this
disable the extension for this game until the shader emission can be fixed.

Related: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6857

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17728>

r600ß/sfn: Don't fall through with VS outputs

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17729>

r600: don't clone the nir shader, just use it

The nir shader passed in with the shader state has to be released by
the backe-end, so don't clone it, just use it.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17729>

r600: Delete possible old NIR variant when translating from TGSI

With shader variants it may happen that we already translated a TGSI
shader for the current selector, so delete the old nir shader if we
already had one.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17729>

r600/sfn: lower the right shader variant

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17729>

r600/sfn: make sure the memory pool is released after shader translation

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17729>

anv: set image_read_without_format NIR option on Vulkan 1.3

VK_KHR_format_feature_flags2 is core and implicitly enabled in 1.3.

Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17442>

radv: Fix out of bounds access in RTPSO creation

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17723>

tu: increase maxPushConstantsSize to 256.

Now there are two paths for push constants.

When it's range is under 128b, we can use shared consts.
When it's over 128b, we can instead do loading data through
regular path, which is same as the previous way.

Now we can satisfy emulations like vkd3d that requires 256b for
its root signatures and we think it fairly maps to push constants
rather than inline uniform blocks that requires one indirection.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15503>

turnip: Change to use shared consts for PushConstants

Follow the way blob is doing for PushConstants though it supports only
128b, same as previous.

v1. Rename tu_push_constant_range.count into dwords to redue confusion.
( Danylo Piliaiev <dpiliaiev@igalia.com> )

v2. Enable shared constants only if necessary.

v3. Merge the two draw states TU_DRAW_STATE_SHADER_GEOM_CONST and
TU_DRAW_STATE_FS_CONST as shared constants are used.

Note that this leaves tu_push_constant_range in tu_shader so we could
use it again in the following patch.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15503>

turnip: clean up unused parameters for user consts.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15503>

ir3: handle shared consts.

Adds a shared consts base offset and a size of it(dwords) to ir3_compiler
since they might be depending on gpu generations. (Danylo Piliaiev <dpiliaiev@igalia.com> )

Adds a flag to present whether shared consts are enabled to
ir3_shader_options and then it sets to ir3_const_state when creating
an ir3 variant. Although this state is not per-shader state, this is
necessary when figureing out real constlens.

v1. Define a hw quirk for geometry shared const files and use it when
calculating const length.

v2. Don't hardcode when calculating a safe const length.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15503>

ir3: change maximum size of const files.

According to the observation on a630/a650/a660, max_const_pipeline has
to be 512 when all geometry stages are present. Otherwise a gpu hang
happens. Acoordingly maximum safe size for each stage should be under
(max_const_pipeline / 5 (stages)).

Only when VS and FS stages are present, the limit is 640.

v1. Align max_const_safe to 4 vec4's.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15503>

gallium: add pipe_blit_info::dst_sample, update u_blitter to write that sample

A new radeonsi blit test will use it to upload different sample values
into MSAA textures.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17694>

radeonsi: check last_dirty_buf_counter and dirty_tex_counter

Check both counters in draw and compute, otherwise compute dispatches may
miss buffers invalidation.
This fixes the test case from https://gitlab.freedesktop.org/mesa/piglit/-/merge_requests/702
(both with and without GALLIUM_THREAD=0).

cc: mesa-stable

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17394>

gallivm: push LLVM version guard into assert

The asserts that check the pointer element type can't be used on LLVM >= 15.
Instead of using precompiler #if, use boolean shortcut in assert.

Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17650>

gallivm: add lp_build_struct_get() variants that take the LLVM type

This is needed for LLVM-15 opaque pointers.
The new variants taking the type are named with the suffix "2", using
the same naming pattern LLVM (e.g. LLVMBuildGEP2 vs. LLVMBuildGEP).

Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17650>

gallivm: refactor a bit the cache access in view of LLVM opaque pointers

LLVM 15 requires transition to opaque pointers; factorize a bit the cache
memthods to help this transition.

Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17650>

gallivm: fix a few llvm non-opaque pointers

As LLVM 15 transitions to opaque pointers, we need to update
the deprecated methods dealing with non-opaque pointers.

Reviewed-by: Brian Paul <brianp@vmware.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17650>

dzn: Retrieve validator version

For platforms where there's no validator available, leave the field zero-initialized
to let the DXIL backend choose whatever target validator version it wants.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

d3d12: Retrieve validator version

For platforms where there's no validator available, leave the field zero-initialized
to let the DXIL backend choose whatever target validator version it wants.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Blacklist DXIL validator 1.6 from 20348 SDK

This version claims to support validator version 1.6, but doesn't
actually have the 1.6 changes (PSV v2, PSV resource v1, barycentrics).

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Add a getter for a validator version from DXIL.dll

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Support DXIL validator 1.7

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Support DXIL validator 1.6

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: DXIL validator 1.6 uses a new PSV resource struct

This is a huge pain because it's an array, meaning that accessing
an entry in the array now depends on the validator version to use
the right element stride.

We could always just store the v1 and downconvert if needed... but
this isn't *that* bad that I felt I had to do it that way.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: DXIL validator 1.6 uses a new PSV struct version

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Set the barycentrics flag for attribute_at_vertex

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: DXIL validator 1.6 counts UAVs differently

Instead of counting the number of UAV arrays, it counts the
number of actual UAVs declared. This is more correct, but we
need to do the same accounting to set the 64 UAVs flag.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Correctly compute dynamic indexing I/O masks

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Compute correct usage masks

This starts actually updating the always-read/never-written
masks while processing the shader. Note that we follow DXC's
lead here and treat "always read" as "sometimes read."

This isn't strictly required, but might help drivers out.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Support DXIL validator version 1.5

This version of the validator starts adding usage masks into
the DXIL, which then are expected to match the PSV and signature
data. The usage masks are "correct" meaning that the never-writes
mask no longer includes bits outside of components 0-3.

A future change will actually compute useful masks.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Lower I/O to scalar

A future change will start computing component masks while
processing I/O instructions, and only having to compute
a mask for one component per instruction simplifies things.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Ensure 4-component position writes via NIR

We're about to lower I/O to scalar, which means we'll end up with
multiple writes to position, and none of them has enough info to
fill in the blanks.

This causes a test that previously crashed on WARP (due to
StoreOutput with an undef not being handled) to fail more
gracefully - but that failure means that the test spends
forever just outputting errors, so explicitly skip it.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Split signature processing into two parts

First, preprocess the signatures, strictly based on the variables
in the nir shader. Then, later, after the actual shader contents
have been processed, we emit the metadata.

This lets shader processing rely on the pre-processed data (e.g.
the row -> ID mapping needed for large VS inputs) while also allowing
the signature data to rely on data gathered during the shader traversal
(e.g. which components are actually used).

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Emit metadata based on long-lived signature data

Instead of using the short-lived semantic structure (that's used to
fill out the long-lived signature and PSV data), use the long-lived
ones. This is staging so we can hold off on emitting the metadata
until later.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Add a max validator version

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

microsoft/compiler: Add a max shader model option

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

nir_lower_io_to_scalar: Support arrayed (per-vertex) I/O

Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17603>

r600: don't dump shader info to files on debug

This was useful to bring up the NIR backend, but is not needed
anymore.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: Fix streamout with non-zero component write

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: Fix used of local shared write mask

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: Fix emmision of LDS instructions

Assert to catch more possible erros with LDS fecth/read
groups.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: Fix the kcache failure handling

Instead of starting a new block when the kcache handling failed,
try to continue scheduling instructions until kcache allocation
fails for all ready instruction.
With that we avoid a CF split withing an LDS fetch/read group.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: count LDS queue pop reads separately in assembler

Otherwise the check whether the fetches and reads are balanced
could fail.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: lower undef to zero

It's what the backend would do anyway, so let's do it in nir and
give the optimizer some chance to profit from possible improvements.

Fixes a bad shader with "The Raven Remastered"

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: fix storing shader output array locations

When lowering gl_Clipertex the driver_location may no longer correspond
to the array index, so fill the array by counting the array index up
according to outputs that need to be handled by the state setup.

Fixes: 3340c7ce359252ad09b3e4d338837944482fb248
    r600/sfn: lower CLIPVERTEX to clip planes

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: clean up some commented out code

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17678>

r600/sfn: Initialize TestShaderFromString member m_instr_factory.

Fix defect reported by Coverity Scan.

Uninitialized pointer field (UNINIT_CTOR)
uninit_member: Non-static class member m_instr_factory is not
initialized in this constructor nor in any functions that it calls.

Fixes: 79ca456b483 ("r600/sfn: rewrite NIR backend")
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/17584>

venus: add support for VK_KHR_dynamic_rendering

Signed-off-by: Ryan Neph <ryanneph@google.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17668>

venus: refactor VkCommandBufferBeginInfo fixups to function

Signed-off-by: Ryan Neph <ryanneph@google.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17668>

panfrost: Allow NULL streamout targets

Fixes Piglit test
arb_enhanced_layouts-transform-feedback-layout-qualifiers.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17447>

panfrost: Don't unbind recently bound streamout targets

When unbinding extra targets, start after the last new target.

Fixes: 5ff79735601 ("panfrost: Import streamout data structures")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17447>

panfrost: Add a debug option for checking overflows on pool uploads

PAN_MESA_DEBUG=overflow will place objects as close as possible to a
protected region at the end of the buffer, so that overflows segfault.

Caught the bugs in all four of the preceding commits.

v2: memset the BO to 0xbb to catch code expecting zeroed allocations.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17447>

panfrost: Correctly calculate prefetch suppresion varying index

The xfb_base is a base index, it makes no sense to multiply that with
the number of streamout targets. Use addition instead to fix a buffer
overflow.

Fixes: 557633b142e ("panfrost: Suppress Bifrost prefetching")
Reported-by: Luc Ma <onion0709@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17447>

panfrost: Clear the GENERAL varying buffer field if unused

Otherwise the indirect draw shader can read uninitialised data for the
stride, and the position varying buffer may be outside the heap BO.

The next commit fixes a bug that masked this one.

Fixes: 2e6d94c198e ("panfrost: Add helpers to support indirect draws")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17447>

panfrost: Emit the correct number of attributes

create_vertex_elements_state is sometimes called with a too large
num_elements argument, for example with util_blitter, which causes a
buffer overflow.

There is no documentation to forbid this practice, so don't rely on
so->num_elements being correct and instead use the vertex shader
attribute count, which matches the value used to allocate the
descriptors.

Use attributes_read_count rather than attribute_count because the
latter also includes images and PAN_VERTEX_ID/PAN_INSTANCE_ID.

Fixes: 76de3e691c6 ("panfrost: Merge attribute packing routines")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17447>

panfrost: Only emit images when they are present

nr_images is the trigger for allocating double the number of buffers
for attributes. When there are no images, there is not always enough
space for ALIGN_POT(k, 2) to not move k out of bounds, so don't
execute the line in that case.

Fixes: dc85f65e059 ("panfrost: emit shader image attribute descriptors")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17447>

zink: remove deqp fails for lavapipe

these are all broken tests, so ignore whatever results there were for now

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17721>

lavapipe: propagate xfb info for pipeline library rasterization stages

Fixes: 202bbedc55a ("lavapipe: streamline xfb shader setup")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17716>

spirv: Fix array length of buffers larger than INT32_MAX.

Like 90a8fb03556e.

fossil-db results:

All Skylake and newer Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 141442369 -> 141442363 (-0.0%)
Instructions helped: 1

Cycles in all programs: 9099270231 -> 9099270187 (-0.0%)
Cycles helped: 1

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17637>

ci/crocus: Fix the commented YML for blender-demo-cube_diorama.

Fixes: c0930b552d01 ("ci/crocus: Disable the blender trace.")
Acked-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17637>

intel/vec4: Set lower_usub_sat

Reviewed-by: Emma Anholt <emma@anholt.net>
Closes: #6900
Fixes: 90a8fb03 ("nir/lower_io: Fix array length of buffers larger than INT32_MAX.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17637>

nir: Split usub_sat lowering flag from uadd_sat.

Intel vec4 would like to do uadd_sat, but use lowering for usub_sat.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17637>

anv: disable dev.i915.perf_stream_paranoid=0 warning

this is spammed constantly even if ANV isn't used

fixes #6731

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17661>

iris: Use fill_surface_states for compressed resources

In iris_create_surface, use the fill_surface_states helper function instead of
an open-coded solution for compressed resources.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17598>

iris: Don't leak compressed resources in iris_create_surface

Before this patch, we were leaking compressed resources in iris_create_surface.
Specifically, when we failed to create an uncompressed ISL surface and view for
a compressed resource, we didn't unreference the resource pointer we referenced
into the pipe_surface.

Fix this by delaying the pipe_surface initialization code to after attempting
to create the uncompressed surface and view.

Cc: 22.1 <mesa-stable>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17598>

iris: Don't leak surface states for compressed resources

Before this patch, we were leaking surface states in iris_create_surface.
Specifically, when we failed to create an uncompressed ISL surface and view for
a compressed resource, we didn't free surface states we allocated for it.

Fix this by attempting to create the uncompressed surface and view before we
allocate the surface states.

Cc: 22.1 <mesa-stable>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17598>

vk/util: simplify extensions gen code

Signed-off-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17713>

vk/util: handful of pythonic cleanups

No functional changes.

Signed-off-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17713>

radv/rt: simplify lower_rt_instructions()

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17301>

radv/rt: fix nir_builder cursor in lower_rt_instructions()

Fixes: 207ce6d658ac6d8f6421a02304b74645ff835e96 ('radv: Add helper to inline shaders into the main shader.')
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17301>

radv/rt: Don't load ClosestHit SBT on every hit, but only once after traversal

Quake II RTX:
Totals from 7 (0.01% of 134913) affected shaders:
CodeSize: 217592 -> 215956 (-0.75%)
Instrs: 39468 -> 39341 (-0.32%)
Latency: 761581 -> 746802 (-1.94%)
InvThroughput: 507721 -> 497870 (-1.94%)
Copies: 4621 -> 4585 (-0.78%)
Branches: 1598 -> 1584 (-0.88%)

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17301>

radv/rt: use stage ID as handle for general and closestHit shaders

This avoids some code duplication and divergence.

Quake II RTX:
Totals from 7 (0.01% of 134913) affected shaders:
CodeSize: 218880 -> 217592 (-0.59%)
Instrs: 39692 -> 39468 (-0.56%)
Latency: 789091 -> 761581 (-3.49%)
InvThroughput: 526061 -> 507721 (-3.49%)
VClause: 1202 -> 1188 (-1.16%)
Copies: 4649 -> 4621 (-0.60%)
Branches: 1605 -> 1598 (-0.44%)

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17301>

radv/rt: use derefs for the traversal stack

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17301>

radv: create RT traversal as separate shader

This will help in future to keep the main shader slim
when we have actual function calls.

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17301>

radv: Only create noop-fs for graphics pipelines

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17301>

anv: implement Wa_14015264727 for DG2

On DG2 we need to flush data cache before fast clear operation.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17218>

iris: implement Wa_14015264727 for DG2

On DG2 we need to flush data cache before fast clear operation.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17218>

d3d12: Remove state tracking implementation details from header

Now that the old state tracking code is removed, implementation details
no longer need to be leaked out of this single source file. Remove structs,
function declarations, 'd3d12_' prefixes, and add static when possible.

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Add a transition flag indicating that state accumulation is needed

Most call sites for transitions will only apply transitions to one or two
resources, and don't need to use the bo set, where each call is guaranteed
to insert the bo, only to walk the set immediately afterwards. Instead, they
can just append the barriers to the dynarray directly and skip the bo set.

Draws and dispatches still use the append approach, to accumulate the full
set of state needed for each subresource for the case where a single
[sub]resource is bound to the pipeline in multiple places.

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Optimize transition_subresource_states that covers a whole resource

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Extract core barrier logic

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Rename bind invalidate options to transition flags

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Swap the remainder of state tracking to new method

Uses a set of d3d12_bo on the context to track which bos are pending
a transition instead of an intrusive linked list, since the bo may
need to be pending on multiple contexts at once.

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Record a state fixup command list when necessary

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Track a global resource state for non-simultaneous-access resources

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Create/free context state entries

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Treat depth/stencil as planar for plane count helper

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Add a context state tracking structure

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Notify contexts about deletion of bos

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>

d3d12: Hold lock when removing resources from residency list

Also, remove them from the list before releasing the ID3D12Resource.

Fixes: 671deb54 ("d3d12: Add residency info to d3d12_bo")
Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17688>