docs: update DRI_PRIME documentation

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24750>

device_select_layer: support DRI_PRIME=n

For feature-parity with OpenGL.
See commit "loader: extend DRI_PRIME to support =N".

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24750>

loader: add DRI_PRIME_DEBUG env var

This makes debugging non-working DRI_PRIME selection much easier.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24750>

loader: extend DRI_PRIME to support =N

DRI_PRIME=1 isn't useful on machines with more than 2 GPUs.

This commit adds support to DRI_PRIME=N syntax meaning: select
the Nth GPU (not counting the default GPU).

So on a 3 GPUs system where drmGetDevices2 returns the following:

   /dev/dri/renderD130 [default]
   /dev/dri/renderD129
   /dev/dri/renderD128

DRI_PRIME=1 would select D129 (as is already the case without this
commit), DRI_PRIME=2 would select D128.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24750>

loader: refactor DRI_PRIME handling code

DRI_PRIME has 3 different meanings and this refactoring
separate the handling of them.
No functional changes.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24750>

glx: drop the 'libGL' log prefix

Suggested by @ajax in !24750.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24750>

ac/spm: enable support for multi-instance counters

This is what RGP expects and this will give us more fine grained
results given that all shader engines/shader arrays etc would be
sampled.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

ac/spm: move the counter instance to ac_spm_counter_create_info

This will allow us to configure multi-instance counters.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

ac,radv,radeonsi: prepare support for multi-instance SPM generic counters

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

ac,radv,radeonsi: prepare support for multi-instance SPM SQ counters

Each SQG modules can configure up to 16 counters.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

ac/spm: fix number of instances of GL2C

It's a global block, so the number of instances is equal.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

radv,radeonsi: make sure to emit GRBM_GFX_INDEX before SQ select registers

This was missing, but not sure if it was required given that only the
first SE is currently sampled.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

ac/spm: select correct segment type for per-SE blocks

This currently does nothing because only the first instance is used,
but this will be needed for multi-instance.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

ac/spm: use block flags to initialize instance mapping

This simplify this a bit, ideally we would also have a per-SA flag
for performance counters.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

radv: reserve more CS space in SQTT/SPM paths

This will prevent an assertion when SPM will emit more counters.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25240>

compiler/types: Use a string table for builtin type names

This avoids the relocations for each of the builtin type names, allowing
all the builtin data to be loaded in read-only memory.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25191>

compiler/types: Change glsl_type::name to be an uintptr_t

This will allow us later to store builtin names in a different way.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25191>

compiler/types: Use glsl_get_type_name() to access the type name

This will allow us later to store builtin names in a different way.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25191>

compiler/types: Use Python to generate code for builtin types

Will be useful later to generate string tables for the builtin types.

Note we make some extra effort to ensure C++ client code doesn't need to change,
by keeping glsl_type::*_type pointers around.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com> (Python and Meson changes)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25191>

r600/sfn: get rid of the method to get the index mode

Since we always split the index load we can query the value directly
without checking whether it was lowered.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25159>

r600/sfn: Sepeate resource and sampler in texture instructions

Fix the use of resource and sampler and keep separate index registers
for both so that indirect resource and sampler access can be done
independently.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25159>

r600/sfn: Add old address to update_indirect_addr

v2: use UNUSED instead of adding (void)var; (Vitaly Kuzmin)

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25159>

r600/sfn: rename method resource_base to resource_id

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25159>

r600/sfn: Make address split pass obligatory

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25159>

r600: print texture resource index mode separately

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25159>

intel/common: Add sse2_args for 32-bit build when -Dsse2=false was set

Fixes: 543a707b7b8 ("intel/common: Move intel_clflush.h to intel_mem.h/intel_mem.c")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9762
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25249>

vulkan/android: improve vkQueueSignalReleaseImageANDROID

There're two issues with the current implementation:
1. Wait semaphores are implicitly required to be SYNC_FD exportable
2. As a queue command that can further record cmds against the wsi
   image, it currently doesn't account for pending cmds in the queue
   beyond the wait semaphores.

This change fixes both by doing a queue submission in the call with a
SYNC_FD external signal semaphore. However, due to Android wsi not
exposing swapchain to icd, we have to cache the signal semaphore in the
queue, otherwise would have to create/destroy in each present.

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25185>

vulkan: remove a dup entry from vk_image_usage_to_ahb_usage

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25185>

Android.mk: Only link LLVM for radeonsi, not amd_vk

When building for Android, avoid the LLVM dependency if possible.

Cc: mesa-stable
Acked-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Change-Id: Idda03f954b4b5326e23a848e541411b60fcef063
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24007>

Android.mk: Explicitly enable/disable LLVM support

The AMD Vulkan driver uses LLVM by default, but it is possible to build
the driver without the LLVM dependency. In this case, we must explicitly
disable LLVM support, or else meson will die after failing to find LLVM.

The Android build system already knows when to link libLLVM, so forward
that information to meson.

Cc: mesa-stable
Acked-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Change-Id: I7489d3811625b390aaaf2e84e666b4a8d98328b0
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24007>

Android.mk: Allow building only Vulkan drivers

Android bundles ANGLE for implementing OpenGL ES and EGL on top of
Vulkan. When using ANGLE, mesa is only needed for its Vulkan drivers.

Cc: mesa-stable
Acked-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Change-Id: I034a0af52f9216bc5f1322f584bc591c90222327
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24007>

meson: use a single dependency call for lua

This is a newer Meson construct that allows passing multiple names for
the same dependency to one call. One advantage of this is that if the
first call (lua54) fails, it won't immediately fall back to a subproject
and will try the others before falling back.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25251>

ci/zink: drop redundant `MESA_LOADER_DRIVER_OVERRIDE: zink`

This job extends `.zink-anv-test` and `.zink-trace-test`, both of which
extend `.zink-test` which already sets this var.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25248>

mesa: some cleanups for texparam extension checks

Patch removes some redudant _mesa_is_desktop_gl checks, these
extensions are available only in desktop context.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25202>

Revert "ci: disable lima LAVA lab for maintance"

This reverts commit 079604e41efe001ab560fa68850f54475589f801.

Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25242>

ci/farm-rules: re-add "run every container and build job when a farm gets re-enabled"

Turns out I was right the first time, I was just missing that this
should only be done in MRs, which @daniels added in 16527f6ffd0129f2f6e1.

Without this, a "farm re-enable" pipeline will run all the jobs for that
farm, but will have none of the container & build jobs to support these
tests.

This reverts commit 1c3097225a4af294720e111eab3f58d590b85c8c.

Fixes: 1c3097225a4af294720e ("ci: don't run everything just because a farm gets re-enabled")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25243>

ci/farm-rules: rename .disable-farm-mr-rules to make it clear it's only about MRs

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25243>

freedreno/decode: Fix printing chip-id

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25231>

freedreno/layout: Handle 565/etc MSAA special case

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25231>

freedreno: Fix crash with debug msgs enabled

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25231>

zink: use null attachments for null attachments with dynamic render

using dummy surfaces here is dumb

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25121>

zink: propagate rp_tc_info_updated across unordered blits

this otherwise breaks rp optimizing

Fixes: 3a9f7d70383 ("zink: implement unordered u_blitter calls")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25121>

zink: use much bigger dummy surfaces

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25121>

tu: handle unused color attachments without crashing

if the format is not set then this attachment must be ignored

Fixes: 97da0a77341 ("tu: Rewrite to use common Vulkan dynamic state")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25121>

lavapipe: EXT_load_store_op_none

free real estate

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25213>

tu: Actually emit patchpoint for viewports with FDM

At some point I intended to track whether the pipeline had FDM enabled
and set a dirty flag, but I realized that wasn't necessary. However I
forgot to remove the useless tu_cmd_state::pipeline_has_fdm and the
viewport/scissor emission code was still using it. Fix that.

Fixes: 97da0a77341 ("tu: Rewrite to use common Vulkan dynamic state")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25225>

tu: Fix assert in FDM state emission

Making the sub_cs not writeable switches the BO we're emitting to, which
causes an assert failure in tu_cs_end_sub_stream() because it looks like
we have a mismatched start/end. Just make it not writeable later.

Fixes: 97da0a77341 ("tu: Rewrite to use common Vulkan dynamic state")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25225>

tu: Fix tu6_emit_*_fdm size call

cs is not initialized until the line after this.

Fixes: 95104707f18 ("tu: Basic a7xx support")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25225>

tu: Fix per-view viewport state propagation

We no longer use viewport::per_view_viewport, it's part of the program
state instead, but we forgot to delete it and rewrite all users.

Fixes: 97da0a77341 ("tu: Rewrite to use common Vulkan dynamic state")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25225>

ci: document which image tags need to be bumped when updating {alpine,debian,fedora}/x86_64

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25120>

ci: document which image tags need to be bumped when updating piglit

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25120>

ci: Fix pre-merge pipelines with no code changes

Make sure we don't end up with manual container jobs in MR pipelines,
because they'll never start. This cleans up the container and build
rules so always 'do the right thing' on MRs (container+build if
required, nothing if not) and user branches (all manual).

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25219>

intel/fs: Check if the whole ubo load range is in the push const range

Before this, we were checking only the beginning of the ubo range, so
partially overlapping loads were trying to load undefined data.

Fixes: b2da1238 ("i965: Use pushed UBO data in the scalar backend.")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9748
Signed-off-by: Sviatoslav Peleshko <sviatoslav.peleshko@globallogic.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25111>

panvk: Add Android ICD loader entry point

Enables offscreen rendering on Android.

Signed-off-by: Roman Stratiienko <r.stratiienko@gmail.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25194>

lavapipe: clamp cache uuid size

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25232>

zink: flag db maps as unsynchronized

these otherwise can be created using the wrong slab

Fixes: 98c411a2a19 ("zink: add PERSISTENT for db buffer maps")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25227>

aux/tc: don't use pipe_buffer_create_with_data() for rp-optimized subdata

this function doesn't use the correct tc map flags, which causes
drivers to (potentially) use the wrong slab for transfer allocation
and then crash

Fixes: 51ad269198e ("aux/tc: handle stride mismatch during rp-optimized subdata")

Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25227>

aux/tc: fix rp info handling around tc_sync calls

if a set_framebuffer_state call has occurred but no draws have been
triggered, the rp info must be preserved for the driver to (eventually)
use after the sync

Fixes: 07017aa137b ("util/tc: implement renderpass tracking")

Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25227>

compiler/types: Fix array name dimension flipping for unsized arrays

Rewrite the code to use linear_asprintf and always flip the
dimensions in place if the element type is an array.  The new
code will now (correctly) flip even in the case of unsized arrays.

The flipping is done by swapping the ranges [a, b) and [b, c), as
shown below, with element type int[...] and an array of length 4.

```
           +--------------- a: first bracket in the name
           |    +---------- b: end of the element name
           |    |  +------- c: end of the array name
           |    |  |
        int[...][4]$

will be transformed into

        int[4][...]$
```

Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23278>

glsl: Annotate _mesa_glsl_error() with PRINTFLIKE

And fix all the errors it found.

Note that for the unsized array error, we will print the
toplevel type -- so that the fact that an inner array is
unsized can be seen.

Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25200>

freedreno: Build drm subdir before perfcntrs, which uses it.

Fixes: f8ac56895509 ("freedreno: Only add drm/computerator when system_has_kms_drm")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25229>

ci: add locked flag to bindgen-cli on x86_64_build.sh

since the dependencies were not locked, they got updated and generating
a new container is throwing errors like the following:

error: failed to compile `bindgen-cli v0.62.0`, intermediate artifacts can be found at `/tmp/cargo-installcP54m7`
Caused by:
  package `memchr v2.6.3` cannot be built because it requires rustc 1.61 or newer, while the currently active rustc version is 1.60.0

rust packages have Cargo.lock file from when they were released, so add
--locked flag to use it.

Signed-off-by: Helen Koike <helen.koike@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25226>

intel/fs: Merge copy prop dataflow loops

This is kept as a separate commit because the change looks like a lot
more than it it. The order of the two loops is swapped, then the two
loops are merged.

Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Use rb_tree for copy prop dataflow

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Use rb_tree to store ACP entries by destination

Using a single data structure seems better. There's no appreciable
performance change. On batman_arkham_city_goty.foz, the difference
reported was 0.48%±0.36% (n=20). Several commits in the MR, including
some that should have no effect at all, reported similar changes. I
attribute this primarily changing of loop alignments and similar.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Use rb_tree to store ACP entries by source

On batman_arkham_city_goty.foz, this improves fossil-db time by
-3.83%±0.24% (n=20). This fossil takes the longest time of any in my
database.

v2: Add some comments for cmp_entry_src_entry_src and
cmp_entry_src_nr. Suggested by Ken.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Encapsulate per-block ACP in a structure

This simplifies some later changes.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Make opt_copy_propagation_local file private

This annoyed me durning development of this MR. Every time I changed the
parameters to this internal function, I had to modify a public header
file... and trigger a much large rebuild.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Simplify check in can_propagate_from

The larger predicate here already requires that inst->opcode must be
BRW_OPCODE_MOV, so it can't BRW_OPCODE_SEL. With that removed, the
other simplifications are pretty straight forward.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Don't loop in try_constant_propagate

The caller already loops over the sources. This means that the caller
must loop over the sources in reverse because constant propagation
prefers to propagate into the last sources first.

The shader-db and fossil-db changes (below) are all due to SEL
instructions. Changing the order sources are visited changes whether a
SEL with two immediate sources is

    (+f0.0) sel     g12    IMM_A    IMM_B

or

    (-f0.0) sel     g12    IMM_B    IMM_A

The ordering of the sources affects the order the constant combining
encounters the values, and the determines which value is "combined"
and which value remains an immediate.

This affects the results by luck. If there are two instructions:

    (+f0.0) sel     g12    IMM_A    IMM_B
    (+f0.0) sel     g13    IMM_A    IMM_C

Picking IMM_A is advantageous over picking IMM_B and IMM_C. Since the
selection algorithm in constant combining is greedy, this case
requires the algorithm see the values in just the right order for the
right thing to happen.

v2: Rebase on many, many changes. Move instruction source fixup
reordering out or try_constant_propagate.

v3: Rebase on !7698.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Move src.file checks out of try_constant_propagate and try_copy_propagate

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Make try_constant_propagate and try_copy_propagate file private

This annoyed me durning development of this MR. Every time I changed the
parameters to this internal function, I had to modify a public header
file... and trigger a much large rebuild.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Don't try to copy propagate into a source again after progress is made

If the linked list structure used depended on the list head to know when
to terminate, this would be a pretty serious bug. If try_constant_propage
or try_copy_propagate make progress, inst->src[i].nr will change. This
results in the foreach_in_list using a different list header on later
iterations of the loop.

This causes two shaders in shader-db and 9 shaders in fossil-db to
change. Looking at the code changes, these are cases where there was a
copy of a copy that gets propagated. The part that confuses me is the
VGRF numbers involved should **not** hash to the same bucket, so it
should be impossible to find the original source from the intermediate
VGRF.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/fs: Don't continue fixed point iteration just because liveout changes

Unless the change in liveout also causes livein to change, updates to
liveout cannot have any global effect. Changes to livein already flag
additional interation.

I had additional changes in this area that didn't pan out. While working
on those change, I was a little confused about this bit of code. It's
unnecessary, so it's better to delete it.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

util/rb-tree: Inline rb_tree_init

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

util/rb-tree: Work around C++'s dislike of offsetof

This is the same technique used in src/compiler/glsl/list.h.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

glsl/list: Clean up an inappropriate comment

:facepalm:

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25091>

intel/genxml: Auto-import genxml files using genxml_import.py

$ src/intel/genxml/genxml_import.py --import

This can be reversed with:

$ src/intel/genxml/genxml_import.py --flatten

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Start Xe2 support

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/decoder: Implement support for importing genxml

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/decoder: ralloc_steal() values from spec context for fields and enums

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Add genxml_import.py script

This script can:
 * validate that genxml files do not duplicate imported items
 * add imports to genxml files and optimize the file by dropping
   duplicate items
 * reverse the import operation by flattening genxml files

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Add GenXml.flatten_xml() method

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Add GenXml.add_xml_imports method

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Drop assertion to allow for importing

For example, gen11.xml will import the HEVC_ARBITRATION_PRIORITY
struct from gen9.xml.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Add GenXml.optimize_xml_import()

This function drops duplicated items from a genxml file when they are
equivalent to the same item imported from another genxml file.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Add all xml files as pack dependencies

Since the output can now depend on other imported xml files, we need
to add them all as dependencies to ensure that if any xml file is
changed, then all pack files are rebuilt.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Add support for excluding items when importing

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

intel/genxml: Support importing from another genxml file

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20593>

glx: Increment dpy->request before issuing an error that had no request

This ensures the sequence number is unique and recent enough for callers
of `glXQueryDrawable` using `XNextRequest` to selectively trap errors.
The same approach is already used in `glXCreateContextAttribsARB`.

Suggested-by: Sebastian Keller <skeller@gnome.org>
Related-to: https://gitlab.gnome.org/GNOME/mutter/-/issues/3007
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25173>

radv: Remove ray tracing shader module identifier skips

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25142>

radv/rt: Handle stages without nir properly

Fixes: e039e3cd768bef6c21255aeef2b11cedf1ab29b9 ('radv/rt: Store NIR shaders separately')
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25142>

radv: Update navi21 llvm fails

Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25010>

radv: Don't advertise features requiring PS epilogs with LLVM

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25010>

ac/llvm: Use float types for float atomics

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25010>

ac/llvm: Use the correct return type for uadd_carry and usub_borrow

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25010>

ac/llvm: Fix typed loads with 16bit formats

For some reason, LLVM can't handle those. Emit a 32bit load and type
conversion instead,

Fixes: 22ca8c8 ("ac/llvm: Implement typed buffer load intrinsic.")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25010>

radv/ci: Improve ray tracing skips

I didn't know they were regexes. This also excludes all "1048576" tests.
They build an acceleration structure with 1 primitive 1048576 times
which only warms up the Valve farm and doesn't accomplish anything else.

Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24720>

radv: Perform multiple sorts in parallel

This was the last part that didn't scale with multiple infos. Reducing
the amount of barriers in this case improves DOOM Eternal performance by
50%. (Running with low resolution)

Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24720>

radv/radix_sort: Vendor the radix sort dispatch code

This needs to be done so we can optimize it for occpuancy when building
multiple acceleration structures in parallel. Changes to the original
code:

- Change // to /* */
- clang-format
- Replace vkCmd calls with calls to the driver entrypoints
- Add a light weight info struct
- Use radv_fill_buffer directly

Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24720>