mesa: replace ALLOC_PRIMS with persistently-allocated context arrays

This is cleaner.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>

radv: Build acceleration structures using BVH IR

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18769>

radv: Create pipelines for conversion meta shaders

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18769>

radv: Add conversion shader for leaf nodes

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18769>

radv: Add conversion shader for internal nodes

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18769>

radv: Add BVH IR types

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18769>

radv: Split CmdBuildAccelerationStructuresKHR into different functions

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18769>

radv: Rename internal node shader to lbvh_internal

Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18769>

zink: only emit runtime array for ssbos if present

it's impossible to accurately determine whether an op will use the base,
sized array or the unsized array due to dynamic indexing, and thus it's
impossible to not violate spec by emitting both arrays at offset=0 so they
can overlap

so instead, just emit the runtime array and blast all bounds checking in
vk drivers into the void

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: remove color_write_missing driver workaround

this is no longer used

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: allow reordered clear_buffer calls

rare to hit, but no point in not promoting them when possible

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: avoid overflow when clamping bufferviews

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: name bo variables using bitsize

debugging++

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: fix handling for ssbos that are just runtime arrays

this used to correctly calculate runtime array size as zero, but since the
switch to glsl_get_explicit_size(), zero can no longer be returned

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: fix sparse queue creation

to use an explicit sparse queue, it has to actually be created along with
the device

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: always unset var->data.explicit_xfb_buffer

ensure that this isn't accidentally triggering for non-xfb shaders

Fixes: 6d40db84c9d ("zink: handle direct xfb output from output variables")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: free vertex_state_cache on shutdown

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: fix some batch tracking leaks

not leaking over time, just leaking on shutdown

Fixes: 83c76b8efbf ("zink: rework batch tracking for resources")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: add asserts to verify optimal key state

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

zink: copy optimal key always

this is going away in the future, but for now it's annoying

Fixes: 7336580408a ("zink: unspaghettify some program update code")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>

iris: Ignore aux for copy_region source if there's no unresolved color

This mirrors the behavior of iris_resource_texture_aux_usage(), which
bypasses the aux metadata when there's no advantage to using it.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18790>

lavapipe: propagate shader access info across libraries

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18827>

lavapipe: set writemask for compute shader buffers

I don't know how this wasn't being set?

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18827>

d3d12: Replace size_t with uint64_t in d3d12_video_encoder definitions for cross-plat compatibility

Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18821>

zink: unspaghettify some program update code

the initial implementation of optimal_keys was added inline, but really
it's an entirely different codepath. by separating these out, it makes the
code more readable, and it also allows for slightly better optimization of
of the optimal_keys codepath

~4-5% improvement for drawoverhead -test 7

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: be even more granular with optimal_key program updates

since the bits of each key are easily and efficiently comparable,
the draw-time updating here can be made extremely granular to
update exactly the stages that have changed since the last
time the program was used, further reducing overhead instead of
updating every possible key value

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: simplify bit tracking in zink_gfx_program_update()

much nicer now that gfx dirty bits are isolated

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: split dirty_shader_stages for gfx and compute

this simplifies/clarifies a lot of related code

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: use optimal_keys to massively reduce size of pipeline cache comparisons

since the entire pipeline module key is a uint32_t, this value can be used
for comparisons instead of comparing every shader module

ideally in the future more drivers will support the required features for
optimal_keys, allowing the other variant functions to be deleted

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: add 'optimal_keys' handling for shader keys

if it's known on init that there are no variants for nonseamless cubes,
or inlined values, or decomposed vertex attrs, then shader keys can be
compressed more optimally to reduce the work needed on program updates

more importantly, this reduces the total hash value for all the shader
modules to a single uint32_t (technically 24 bits), which is much better
than having to manage and incrementally add all the separte module hashes

...but for now using this is incompatible with gpl, so disable that

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: add some padding to small shader keys for byte-alignment

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: ALWAYS_INLINE bind_gfx_stage

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: move to AoS for gfx program shader modules

this matches up better with the actual usage; the zink_shader_module
is still stored in the shader cache

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: copy the non-dirty shader stages when updating gfx program

typo during refactoring

Fixes: d8455e5ebfa ("zink: change u_foreach_bit to regular for loop in zink_gfx_program_update")

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: handle inline uniforms during first-time shader generation

this is unlikely but possible

Fixes: a0e69e7601a ("zink: split out first-time shader module creation")

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: remove a bunch of casts from shader key inlines

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: wrap zink_shader_key_base access for nonseamless cubemap use

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: check on init whether decomposed attrs are needed

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

zink: add a bool flag for decomposed vertex elements

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18786>

ci/crocus: update ci results

Seems we haven't been running the Crocus CI much lately, and some things
has changed. The new failures needs to be investegated and fixed, but
let's update the results for now.

Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18822>

aco: mostly implement FS input loads on GFX11

Quad-divergent CF and vertex selection doesn't work, but should at least
prevent crashes.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco/tests: add GFX11 assembly tests

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: omit read-only memory_sync_info when printing

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: add VINTERP instruction format

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: add LDSDIR instruction format

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: limit GFX11 to 128 VGPRs for now

See https://reviews.llvm.org/D128054

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: update assembler for GFX11

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco/gfx11: don't use more than 1 NSA dword

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: add GFX11 opcode numbers

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: fix assembly of MUBUF-to-LDS loads

These have an implicit m0 read and don't write VGPRs.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: add reg() helper to assembler

SGPR encoding is slightly different on GFX11.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

aco: rename Interp_instruction to VINTRP_instruction

These is clearer since GFX11 adds another interpolation format.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17333>

radeonsi/vcn: support async

enable async for vcn

Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18781>

aco: Fixes compiling error about char8_t with c++20

The error is:
../mesa/src/amd/compiler/aco_register_allocation.cpp:382:7: error: no matching function for call to 'printf'
      printf(u8"☐");

Fixes: 209a89e51d1 ("aco: Convert to use u8 literal for Unicode character to fixes msvc warning")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7318

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Marcus Seyfarth <m.seyfarth@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18796>

radv/rt: inline radv_rt_pipeline_create_() helper into radv_rt_pipeline_create()

This saves some re-allocation and will help in future with RT shader functions.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18755>

radv/rt: create separate radv_rt_pipeline struct

inherited from radv_compute_pipeline to contain all RT-related information.
This will make it easier to transition to RT shader functions.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18755>

pvr: Fix assignment to ISP back_a.sref

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <karmjit.mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18778>

radv: use direct access to last_vgt_api_stage_locs for sgpr emission

radv_lookup_user_sgpr is heavy, stop using it

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18808>

radv: store pointer to sgprs for last vertex stage

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18808>

nir: use nir_shader_instructions_pass in nir_split_per_member_structs

Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
  is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_split_var_copies

No functional changes.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_samplers

No functional changes.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_interpolation

Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
  is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_metadata_none instead of its value

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_frexp

Changes:
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_fb_read

Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
  is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_drawpixels

Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
  is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_clip_halfz

Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
  is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_clip_disable

Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
  is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_clamp_color_outputs

Changes:
- removal of lower_state (not needed anymore)
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
  is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_64bit_phis

No functional changes.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_bool_to_int32

No functional changes.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_bool_to_float

Changes:
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_bool_to_bitsize

Changes:
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

nir: use nir_shader_instructions_pass in nir_lower_alu

Changes:
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress
- only metadata of the current function is invalidated (invalidation on
  one function was leaking to successive functions because "progress"
  was not reset)

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

glsl: use nir_shader_instructions_pass in gl_nir_lower_samplers_as_deref

No functional changes.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

glsl: use nir_shader_instructions_pass in gl_nir_lower_images

No functional changes.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

glsl: use nir_shader_instructions_pass in gl_nir_lower_atomics

Changes:
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

crocus: remove invalid nir_metadata_preserve from crocus_fix_edge_flags

nir_metadata_preserve(NULL, ...) doesn't make any sense.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

iris: remove invalid nir_metadata_preserve from iris_fix_edge_flags

nir_metadata_preserve(NULL, ...) doesn't make any sense.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

radv: use nir_shader_instructions_pass in radv_nir_lower_ycbcr_textures

Changes:
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
  make progress

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>

radv: ALWAYS_INLINE radv_is_streamout_enabled()

v2 by Timur Kristóf:
- Use ALWAYS_INLINE instead of just inline.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18807>

radv: ALWAYS_INLINE radv_flush_descriptors

B I G  P E R F

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18807>

radv: ALWAYS_INLINE radv_after_draw

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18807>

radv: enable NGG culling unconditionally for GPL but disable it dynamically

When compiling the pre-rasterization stages we don't know the primitive
topology, but we still want to enable NGG culling for performance. To
achieve that, NGG culling is enabled unconditionally when the topology
is unknown and disabled dynamically for points or lines.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18776>

radv: use the maximum number of vertices per primitives for NGG with GPL

The hw will ignore the extra bits when points/lines are drawn, so this
just works as-is.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18776>

radv: determine the last VGT api stage also for GPL

When compiling the pre-rasterization stages, we need to know this.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18776>

radv: remove useless gfx10_ngg_info::enable_vertex_grouping

It's always TRUE and this will simplify future changes.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18776>

radv: allow to build the pre-rasterization stages in a library

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18672>

radv: determine the last VGT api stage from the active_stages bitfield

With GPL, we can get binaries from libs directly, so this would fail.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18672>

radv: import the GS copy shader from a library if present

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18672>

radv: do not free the GS copy shader binary if created from a library

Similar to other shader stages.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18672>

gallivm: fix buffer_type for LLVM 15

Fixes
deqp-vk: /home/airlied/devel/llvm-project/llvm/include/llvm/IR/Instructions.h:961: static llvm::GetElementPtrInst* llvm::GetElementPtrInst::Create(llvm::Type*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, const llvm::Twine&, llvm::Instruction*): Assertion `cast<PointerType>(Ptr->getType()->getScalarType()) ->isOpaqueOrPointeeTypeMatches(PointeeType)' failed.

Fixes: 1b8c4931df9d ("gallivm: fixes for LLVM-15 opaque pointers in lp_bld_jit_types.c")
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18816>

gallivm: handle llvm coroutines for llvm > 15

LLVM 15 changed the coroutine presplit function attribute in

735e6c40b5e9 [Coroutines] Convert coroutine.presplit to enum attr

This needed to be updated in mesa.

Cc: mesa-stable
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18815>

llvmpipe: bind sampler views/images properly.

There is some code that relies on TGSI here, and it has limits.
For now always just bind resources > 31.

Fixes
dEQP-VK.pipeline.pipeline_library.descriptor_limits*

Cc: mesa-stable
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18812>

llvmpipe: memset empty images in the key

This fixes a bunch of valgrind warnings in
dEQP-VK.pipeline.pipeline_library.descriptor_limits.fragment_shader.storage_images_63

Cc: mesa-stable
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18812>

lavapipe: handle view mask layer count properly for begin rendering.

If the view mask is set the layer count should be taken from it,
not from the layerCount.

Fixes: 8a6160a354282 ("lavapipe: VK_KHR_dynamic_rendering")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18811>

lavapipe: take layers into account around render surface creation.

This makes the code more consistent with handling layers, but also
makes sure a new surface is created when the w/h/layers changes.

Fixes: 8a6160a354282 ("lavapipe: VK_KHR_dynamic_rendering")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18811>

llvmpipe: handle different blocksizes for sampler/image views

In the cases where the sampler/image view format has a
different blocksize to the underlying resource, the width
/height need to be scaled.

Previous code to do this, did the rescaling before minification,
apparently scaling needs to happen after minification, this means
it has to happen in the llvm side.

This fixes a bunch of new VK CTS tests
dEQP-VK.image.texel_view_compatible.*

Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18767>

radv: Use scalar layout for BDA references

Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18794>