radv/android: Remove unused variable

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

vulkan: Allow global symbol HMI for Android

Android looks for this symbol when loading HAL modules.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

anv: Add stub for anv_gem_get_tiling() for Android

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

anv: Pass device to setup_gralloc0_usage for error reporting

Otherwise it doesn't compile.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: aba57b11ee8c6d ("anv: support GetSwapchainGrallocUsage2ANDROID for Android")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

meson: Define ANDROID and ANDROID_API_LEVEL when compiling for Android

Set ANDROID_API_LEVEL based on the value we already have and define
ANDROID to make sure we build code paths that are guarded by that.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

turnip: Make tu_android.c compile again

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

mapi: Mark TLS symbols as optional in glapi-symbols.txt

Presence of these depends on whether or not we're using ELF TLS.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

ci: Include enough Android headers to let us compile test EGL

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

gallium/android: Rewrite backtrace helper for android

The previous implementation kept a hashtable of a Backtrace object per
thread.  debug_backtrace_capture is supposed to store a backtrace in
the passed in debug_stack_frame array, but instead overwrote the
per-thread Backtrace object.

This new version works more like the libunwind based capture. We hash
the file and symbol names and store pointers in the debug_stack_frame
struct.  This way debug_backtrace_capture doesn't overwrite previous
captures or allocate memory that needs to be freed.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

util: Add unit test for stack backtrace caputure

First test never fails, but exercises the code and is useful for
manual inspection.  Second test exposes the android implementation
bug.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

util: Move stack debug functions to src/util

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

gallium: Switch u_debug_stack/symbol.c to util/hash_table.h

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

mapi/test: Change type to unsigned for offset

Quiets this warning:

../../master/src/mapi/glapi/tests/check_table.cpp:576:20: error: non-constant-expression cannot be narrowed from type 'unsigned int' to 'int' in initializer list [-Wc++11-narrowing]
   { "glColor3dv", _O(Color3dv) },
                   ^~~~~~~~~~~~

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

egl/android: Remove unused variable

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6112>

amd/llvm: Reorder LLVM headers

LLVM uses __declspec(restrict) which breaks because Mesa define restrict
as __restrict. Move the LLVM headerse up to dodge the macro.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6180>

ci: Use a rootfs tarball for NFS root, instead of a ramdisk (for LAVA)

We anyway depend already on robust network support in the DUTs, and we
can save quite some time this way.

It will also allow us to grow further as we expand coverage.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-By: Rohan Garg <rohan.garg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6114>

radv: set BYPASS_VTX_RATE_COMBINER_GFX103 on GFX 10.3

Based on RadeonSI.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6168>

radv: fix sample shading on GFX 10.3

Based on RadeonSI.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6168>

radv: increase minimum NGG vertex count requirement per workgroup on GFX 10.3

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6168>

radv: do not honor a user-specified pitch on GFX 10.3

According to RadeonSI, it breaks mipmapping.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6168>

nir/lower_tex: handle query lod with nir_lower_tex_packing_16 at lower_tex_packing

packing_16 with floats assumed 1 (shadow) or 4 components. But query
lod operations return 2.

Fixes the following test with v3dv:
dEQP-VK.ycbcr.query.lod.fragment.r8g8b8a8_unorm

Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5863>

radv/llvm: enable VK_KHR_memory_model

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/6063>

ac/nir: fix coherent global loads/stores

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/6063>

ac/nir: consider an image load/store intrinsic's access

ACCESS_COHERENT may be set for a specific load/store in the case of
atomic loads/stores.

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/6063>

radv/aco: enable VK_KHR_memory_model

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/6063>

spirv: set ACCESS_COHERENT for ssbo/global/image atomic load/store

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/6063>

nir/lower_tex: skip lower_tex_packing for the texture samples query

Similar to other skips for texture queries that don't actually sample
the texture and which results are not packed.

We can't use nir_tex_instr_is_query() here to skip the lowering for all
queries since that causes regressions in Piglit. Apparently, we do want
to lower some of the query results. In particularly, the LOD query.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6169>

freedreno/computerator: Set SP_MODE_CONTROL to the same value as vulkan/GL

This gets us consistent hcN access with our drivers, for experimenting.
We don't know what the other bit does yet, but let's not have to debug
that later.

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6179>

freedreno/a6xx: Document the bit for the magic 32bit-uniforms-as-16b mode.

Trying to figure out how uniforms were working, I found that computerator
had different behavior from our GL fragment shaders.  Given that 3xx had
an SP_ bit for this (thanks flto@ for the note), it was a matter of
pasting bits of SP_* setup into computerator until I got the GL behavior.
I named it the same as the a3xx register.

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6179>

freedreno/ir3: Fix the type of half-float indirect uniform loads.

We would be making a MOV from a u32, when we should be loading from a
16-bit value.  This likely didn't bite us because we only do mediump in FS
and CS so far, and indirect uniforms are usually in a VS (and usually
highp).

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6179>

nir: Print the constant data size associated with a shader.

We should probably dump the constants, too, but this is useful to me for
now.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6179>

nir: Add a little more docs about NIR's constant_data.

I think everyone trips over "how does this relate to nir_const", and I was
curious if I could redefine the units of the constant_data_size / indirect
offsets.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6179>

nir/lower_amul: Use num_ubos/ssbos instead of recomputing it.

Now that num_ubos is correctly maintained, we can just trust it.  Fixes an
assertion failure in freedreno I triggered on
dEQP-GLES31.functional.ubo.random.all_per_block_buffers.1 for reasons I
don't really understand.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6179>

turnip: Only include msm_drm in tu_drm.c

We copy the definition for struct drm_msm_gem_submit_bo and flags to
keep the bo list code working for now.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5999>

turnip: Move remaining drm code to tu_drm.c

This moves the semaphore implementation and tu_QueueSubmit to
tu_drm.c, such that that's the only file including xf86drm.h and
msm_drm.h.  This way, the entire kernel interface is contained in
tu_drm.c

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5999>

turnip: Collapse some tu_drm wrappers

These are all internal to tu_drm.c, we can skip a couple of
abstraction layers now.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5999>

turnip: Move tu_bo functions to tu_drm.c

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5999>

turnip: Move device enumeration and feature discovery to tu_drm.c

These steps are all drm specific.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5999>

anv: fix allocation of custom border color pool

Turns out that respecting the order of parameters is important.

Reported-by: Michael Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Fixes: 5425968d2e46 ("anv: Implement VK_EXT_custom_border_color")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6175>

aco: replace MADs in isel with FMA on GFX10.3

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/5546>

aco: disable SMEM stores on GFX10.3

These are removed in GFX10.3

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/5546>

aco: update aco_opcodes.py for GFX10.3

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/5546>

aco: implement subgroup shader_clock on GFX10.3

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/5546>

aco: update vgpr_alloc_granule for GFX10.3

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/5546>

aco: fix max_waves_per_simd on Polaris, VegaM and GFX10.3

fossil-db (Polaris):
Totals from 20263 (14.75% of 137414) affected shaders:
SGPRs: 871407 -> 871679 (+0.03%); split: -0.00%, +0.03%
VGPRs: 513828 -> 550028 (+7.05%); split: -1.68%, +8.72%
CodeSize: 18869680 -> 18828148 (-0.22%); split: -0.23%, +0.01%
MaxWaves: 162012 -> 162030 (+0.01%); split: +0.01%, -0.00%
Instrs: 3629172 -> 3618817 (-0.29%); split: -0.30%, +0.02%
Cycles: 15682244 -> 15638244 (-0.28%); split: -0.30%, +0.02%
VMEM: 10675942 -> 10673344 (-0.02%); split: +0.18%, -0.21%
SMEM: 1209717 -> 1206088 (-0.30%); split: +0.03%, -0.33%
VClause: 81780 -> 81227 (-0.68%); split: -0.73%, +0.06%
SClause: 231724 -> 231561 (-0.07%); split: -0.07%, +0.00%
Copies: 187126 -> 180831 (-3.36%); split: -3.62%, +0.26%
Branches: 26841 -> 26837 (-0.01%); split: -0.03%, +0.01%

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/5546>

aco: update bug workarounds for GFX10_3

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/5546>

aco: don't create v_mad_f32 on GFX10.3

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/5546>

aco: fix waitcnt insertion on GFX10.3

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/5546>

panfrost: Implement EXT_multisampled_render_to_texture

Significantly helps WebGL performance with Chromium's OpenGL ES backend.
Also update docs/features.txt

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6161>

panfrost: Add MSAA mode selection field

This field enables MSAA, either writing samples to separate surfaces, to
a single large-bpp surface, or implicitly resolved and to a single
surface.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6161>

docs/features: Add GL_EXT_multisampled_render_to_texture

Currently only a6xx, panfrost added later in series.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Acked-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6161>

radv: print warnings for famous RADV_PERFTEST options that no longer exist

RADV_PERFTEST=aco no longer exists, ACO is the default compiler.
RADV_PERFTEST=llvm is deprecated, use RADV_DEBUG=llvm instead.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5662>

radeon/vcn: Corrected vp9 ref associated data incase of target->codec is NULL

This patch fixes the case where less number of reference surfaces created and destoyed
on need basis. The problem comes when we are refereing old assoiciated data for newly
created target buffer with same address. Here old target buffer destroyed as that
surface is no more used as reference for next frames and when we create a new surface
for the next frame to process we will get the surfaceid and same target address
of destroyed surface.

When new surface/surface->buffer/target ,target->codec is null as we cleared when we
destroy this surface, but per ref_mapping logic, it was taking null associated data
i.e.0 as curr_ref_idx. Hence total reference mapping table goes wrong with wrong data.
Beacuse of this, we have seen corrupted vp9 decoded frames.

Signed-off-by: SureshGuttula <suresh.guttula@amd.corp-partner.google.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5452>

spirv: Propagate explicit layout only in types that need it

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5511>

lima: Fix lima_screen_query_dmabuf_modifiers()

Incorrect implementation has been found during code surfing.
v3d implementation used for reference.

Signed-off-by: Roman Stratiienko <r.stratiienko@gmail.com>
Reviewed-by: Icenowy Zheng <icenowy@aosc.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6076>

tu: Implement VK_EXT_conditional_rendering

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6009>

tu: Reset has_tess after renderpass

Don't force sysmem for render passes after the one that uses
tessellation.

Also, move this into tu_cmd_state as that's where it belongs.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6009>

freedreno: Document draw predication packets

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6009>

tu: Fix DST_INCOHERENT_FLUSH copy/paste error

This was meant to handle incoherent accesses by always flushing them,
but it accidentally checked for the coherent variant instead. As a
result e.g. a vkCmdClearImage() followed by a renderpass using the image
didn't get any flushes, resulting in the same sort of corruption seen
with sysmem renderpass clears. This happened to be exposed via some
tests that used multiview.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6009>

turnip: implement VK_EXT_4444_formats

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6164>

util/format: translate A4R4G4B4_UNORM and A4B4G4R4_UNORM vulkan formats

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6164>

turnip: rework extended formats to allow more extended formats

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6164>

ci: Actually upload trace artifacts to MinIO for baremetal

Baremetal jobs filter the variables they get from .gitlab-ci.yml, and
TRACIE_UPLOAD_TO_MINIO and others weren't being let through.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Fixes: d4ca45eca2a5 ("ci: Upload traces' reference and actual images to MinIO")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6171>

turnip: add support for D32_SFLOAT_S8_UINT

Add support for D32_SFLOAT_S8_UINT, which requires special handling because
it is actually two images.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5537>

ci: Remove kernel module build that slipped in

Some changes unintendedly slipped into an unrelated commit before it was
merged.

This caused kernel modules to be built and installed in the ramdisk,
which caused some devices to fail to boot due to the ramdisk size limit
being surpassed.

These changes weren't in effect until a subsequent commit triggered a
rebuild of the ramdisks.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Fixes: a9560939e07a ("ci: Build-test Panfrost tools")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6167>

ci: Download traces from MinIO in baremetal runs

Now that we have MinIO, we can distribute traces better than by direct
downloads from git.

With a caching MinIO instance local to the DUT, total run times should
be noticeably impacted.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6136>

ci: Upload traces' reference and actual images to MinIO

Now that the devices have sane dates, we can upload to MinIO.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6136>

spirv: Do more complex unwrapping in get_nir_type

The OpenGL flavor of SPIR-V allows for samplers inside structs.  This
means that our simple array-of-array handling isn't sufficient and we
need something substantially more complex for generating NIR types.

Fixes: 14a12b771d0 "spirv: Rework our handling of images and samplers"
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6065>

compiler/types: Add a struct_type_is_packed wrapper

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6065>

freedreno: Fix "Offset of packed bitfield changed" warnings:

Example:

../src/freedreno/ir2/instr-a2xx.h:384:1: note: offset of packed bit-field
‘const_index’ has changed in GCC 4.4
  384 | } instr_fetch_vtx_t;

It's apparently due to bitfields that would cross the width of their type.
Just expand the types of the affected fields so that the compiler quiets
down.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6165>

turnip: delete tu_clear_sysmem_attachments_2d

2D path is using the same hardware as the 3D path, with the advantage of
separate register state, but here it requires WFI and extra cache flushing
and invalidating, so it should be better to just use the 3D path. There are
also some cases where the 3D path would be much faster, since it can clear
multiple attachments at once.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5775>

turnip: fix sysmem CmdClearAttachments 3D fallback breaking GMEM path flush

It was clearing the flush bits, which are used by both GMEM/SYSMEM paths,
but emitting the flushes inside the cond_exec, where they would only run
for the sysmem path.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5775>

nir/deref: Don't try to compare derefs containing casts

One day, we may want copy_prop_vars or other passes to be able to see
through certain types of casts such as when someone casts a uint64_t to
a uvec2.  However, for now we should just avoid casts all together.

Fixes: d8e3edb784d3a "nir/deref: Support casts and ptr_as_array in..."
Tested-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6072>

nir: Remove the old nir_opt_shrink_load.

The old pass only handled intrinsic load_constant, while the new
nir_opt_shrink_vectors handles ALU ops, nir load_consts, along with all
the load intrinsics.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6050>

amd: Swap from nir_opt_shrink_load() to nir_opt_shrink_vectors().

This should do much more trimming than shrink_load, and is a win on i965's
vec4 and nir-to-tgsi.  For scalar backends like this that don't need ALU
shrinking, it still gets more load intrinsics covered.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6050>

i965: Enable vector shrinking in the vec4 backend.

This manages to make some extra vec operations that would turn into movs
go away.

brw shader-db:
total instructions in shared programs: 3895037 -> 3893221 (-0.05%)
total cycles in shared programs: 113832759 -> 113792154 (-0.04%)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6050>

nir: Add a pass to cut the trailing ends of vectors.

Ideally we'd also handle unused middles of vectors and reswizzle ALU-only
uses of it so we could write fewer channels, but that's future work/

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6050>

turnip: workaround for a630 d24_unorm_s8_uint fails

A630 doesn't have the HW format we use to sample stencil, so it needs a
workaround. It also has a bug around the AS_R8G8B8A8 format, which doesn't
work when UBWC is disabled, so use 8_8_8_8_UNORM instead when UBWC is
disabled (using AS_R8G8B8A8 or 8_8_8_8_UNORM should only matter with UBWC)

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5438>

etnaviv: fix nir validation problem

Fixes the following validation problem:
  error: nir_intrinsic_align_offset(instr) < nir_intrinsic_align_mul(instr)

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Tested-by: Lukas F. Hartmann <lukas@mntmn.com>
Acked-by: Lucas Stach <l.stach@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6127>

freedreno/decode: move dependencies up a level

This is mainly for the benefit of automated syncing of changes from mesa
back to envytools, where the same subdir meson.build's are used, but the
toplevel meson.build is different.  In the envytools case, we want these
depenendencies to be required.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6154>

freedreno/registers: install gzip'd register database

The decode tools aren't too useful to install without the xml.  But
since libxml2 can read compressed xml, we'll gzip them for installation.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6154>

freedreno/registers: split header build into subdirs

Instead of building the adreno/foo.xml headers from the toplevel, split
out a subdir().  This fits better with how meson likes things to be
structured.  But it does require fixing a bit about how gen_header.py
resolves imports, ie. it cannot assume the src file is at the root of
the $RNN_PATH.

This is needed for the next patch, to add support for installing the
register database for use with installed tools.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6154>

freedreno/registers: add .gitignore

Testing headergen2 will create .xml.h in the same location as the .xml.
But we don't want this to get accidentially committed.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6154>

freedreno/rnn: rework RNN_DEF_PATH construction

No need for rnn_path.h, just construct the whole thing in meson and pass
via c_args.  Also move where the path is constructed up one level.  This
will be needed for syncing back to envytools, where the path will be
different.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6154>

freedreno/rnn: also look for .xml.gz

libxml2 can load gzip compressed files, so lets look for these too.
This will be useful for installing the register database so that an
installed cffdump/crashdec can use them.  But it isn't too useful
to be able to edit the installed register database, so we can gzip
them to use less disk space.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6154>

radv: Implement VK_EXT_4444_formats

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6160>

anv: Implement VK_EXT_4444_formats

We only support the ARGB format, not the ABGR one.  Fortunately, the
ARGB is the one required by D3D11.

Reviewed-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6158>

vulkan: Update Vulkan XML and headers to 1.2.149

Reviewed-by Joshua Ashton <joshua@froggi.es>

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6158>

u_prim_restart: add inline function for getting restart index based on index size

handy to have this available for drivers to reuse

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6147>

iris: Zero the add-on clear color BO on import

When iris imports an I915_FORMAT_MOD_Y_TILED_GEN12_RC_CCS surface, it
allocates a buffer to hold the indirect clear color. When the import is
complete, iris_resource::aux::clear_color is set to zero but the
indirect buffer is filled with garbage values. This could break certain
texture view use-cases, so zero the allocated buffer to fix those.

Fixes: c19492bcdb9 ("iris: Handle importing aux-enabled surfaces on TGL")
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6092>

ci: Build-test Panfrost tools

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3348
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6157>

zink: implement handling for VK_EXT_calibrated_timestamps

just the extension setup; we need this to handle device timestamp
queries outside of batches

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5913>

zink: store valid timestamp bits onto zink_screen

we need this for converting timestamp ticks to nonoseconds

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5913>

zink: handle VK_EXT_vertex_attribute_divisor setup

this just enables the extension

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5913>

zink: clamp PIPE_SHADER_CAP_MAX_SHADER_BUFFERS to PIPE_MAX_SHADER_BUFFERS

this value gets split between ssbos and abos, so clamping to 8 here causes
a number of tests to fail just because there's not enough buffers available

other gallium drivers return 32 here, so this seems pretty safe

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5913>

zink: implement VK_EXT_robustness2

this adds support for null descriptors, which is necessary for full
compliance with ARB_texture_buffer_object

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5913>

pan/mdg: Test for SSA before chasing addresses

It's possible an SSA value depends on a register; in this case, chasing
the source would result in a crash as the chase helper in NIR asserts
is_ssa. Instead we should check a priori that all the argments are in
fact SSA, bailing otherwise.

In the piglit shader exhibiting this bug (by looping over the index),
bailing on the ishl instruction is -necessary-. This is not merely us
being cowardly to avoid seeing through the registers; indeed, if we
wrote away the ishl instruction, the shift itself would have to be
stored in a load/store register (r26/r27) which would preclude reading
it in the loop, creating a register allocation failure later in the
compile. So this is the correct solution due to the restricted
semantics.

Closes #3286

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reported-by: Icecream95 <ixn@keemail.me>
Fixes: f5401cb8866 ("pan/midgard: Add address analysis framework")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6144>

pan/mdg: Mask spills from texture write

This prevents RA failures the results of reading multiple textures that
require less than 4 channels, as seen in a number of GL 3 WebRender shaders.

Closes: #3342
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reported-by: Icecream95 <ixn@keemail.me>
Tested-by: Icecream95 <ixn@keemail.me>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6144>

swr: Bump maximum 2D texture size to 16kx16k

This may be required by some applications, even though
not all texture formats will be below 2GB limit.

This change also increases the maximum size of render target,
that was inadvertently lowered some time ago.

Reviewed-by: Krzysztof Raszkowski <krzysztof.raszkowski@intel.com>
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6139>

mesa/st: Actually free the driver part of memory objects on destruction.

_mesa_delete_memory_object(ctx, obj) == free(obj) but doesn't free the
part of the gallium driver.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/1206
Fixes: 49f4ecc6777 "mesa/st: start adding memory object support"
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6148>