zink: do a better job conserving locations for packed xfb outputs

if an entire vec4 is exported to xfb, mark it as an explicit xfb buffer
whenever possible to avoid blowing out the location limit

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13632>

drirc: add options for BETA CAE Ansa application.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13364>

drirc: add mesa_extension_override option

This allows specific per-application override.
The existing MESA_EXTENSION_OVERRIDE env variable is kept.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13364>

mesa/init: replace call_once with manual implementation

This will be useful to add parameters to one_time_init().

_MTX_INITIALIZER_NP and Windows don't play nice together,
so I had to keep a call_once() to initialize the mutex.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13364>

mesa: don't use dummy_true for some MESA extensions

Otherwise we can't use MESA_EXTENSION_OVERRIDE, or rather: disabling one
extension using dummy_true will disable all others.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13364>

mesa: print a warning when an extension can't be disabled

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13364>

broadcom/compiler: copy packing when converting add to mul

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13675>

radv: Use MESA_VULKAN_SHADER_STAGES to make room for mesh/task.

Signed-off-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/13440>

llvmpipe: add missing NIR alu-op handling

nir_op_bcsel implemented based on ac_nir_to_llvm.c emit_bcsel function.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13246>

mesa: enable force_direct_glx_context for DiscoveryStudio2020

This app uses Qt4.8 which uses an indirect context to save a GL scene to
a pixmap.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13246>

glx/drirc: add a force_direct_glx_context option

Some applications may request an indirect context but this feature is
disabled by default on Xorg and thus context creation will fail.

This commit adds a drirc setting to force the creation of direct glx
context, regardless of what the app is requesting.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13246>

vbo/dlist: free copied.buffer if no vertices were copied

Other parts of the code asserts that copied.buffer is NULL if there are
no vertices to copy.

Cc: mesa-stable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13246>

driconf: set vblank_mode=0 for viewperf2020

It doesn't disable vsync. Reported internally.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13612>

ac/surface: don't validate DCC settings if DCC isn't possible

Cc: mesa-stable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13550>

radeonsi: fix ps_uses_fbfetch value

si_update_ps_colorbuf0_slot used blitter_running as a way to detect
recursive calls.
Unfortunately this catch too many cases; for instance a backtrace
like:
  #0 si_update_ps_colorbuf0_slot
  #1 si_set_framebuffer_state
  #2 do_blits
  [...]
  #5 si_blit
  #6 si_copy_region_with_blit

Would end-up not updating ps_uses_fbfetch; so if the new fb_state is
something like:
  cbufs = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, zsbuf = 0x55b8987545e0}

We can have ps_uses_fbfetch=true but cbufs[0] = NULL, which causes a
crash later in si_ps_key_update_framebuffer.

This commit fixes intermittent crashes in KHR-GL46.stencil_texturing.functional.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13550>

radeonsi/sdma: fix bogus assert

src can use dcc even for non sdma v5 variants because si_decompress_dcc
is called in si_sdma_copy_image.

Fixes: 46c95047bd7 ("radeonsi: implement si_sdma_copy_image for gfx7+")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13550>

ac/surface: use a less strict condition in is_dcc_supported_by_L2

While Mesa chooses to always use independent_128B_blocks, other drivers
can make different choices.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13550>

radeonsi: treat nir_intrinsic_load_constant as a VMEM operation

This is used by variable indexing of constant arrays, to build
code like this:

   s_add_u32 s6, s6, const_data@rel32@lo+4
   s_addc_u32 s7, s7, const_data@rel32@hi+12
   [...]
   global_load_dword v4, v4, s[6:7

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5118
Fixes: 8288882965b ("radeonsi: set MEM_ORDERED optimally")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13550>

radv: Split off cmd_buffer variant of descriptor set template updates

Assumes cmd_buffer != NULL for this path to eliminate the checks in the generic code.

Benchmarks:

I made a microbenchmark based on Bas' vulkan microbench suite.

https://gitlab.freedesktop.org/bnieuwenhuizen/vulkan_microbench/-/merge_requests/1

In this benchmark, this improves vkDescriptorTemplateUpdate performance consistently by 36%.

-------------------------------------------------------------------
Benchmark                         Time             CPU   Iterations
-------------------------------------------------------------------
DescriptorTemplateUpdate       81.2 ns         81.2 ns      8573169

->

-------------------------------------------------------------------
Benchmark                         Time             CPU   Iterations
-------------------------------------------------------------------
DescriptorTemplateUpdate       52.9 ns         52.9 ns     13306065

Signed-off-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13342>

radv: Split off cmd_buffer variant of descriptor set updates

Assumes cmd_buffer != NULL for this path to eliminate the checks in the generic code.

Signed-off-by: Joshua Ashton <joshua@froggi.es>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13342>

radv: Always inline descriptor writes

Improves performance, see next commit for benchmarks.

Signed-off-by: Joshua Ashton <joshua@froggi.es>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13342>

freedreno/a618: Mark a flaky test that triggers hangcheck.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13659>

freedreno/a5xx: Use the defined names for 2D_BLIT_CNTL regs.

We have definitions for them above, no need to be UNKNOWN about it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13659>

freedreno/a6xx: Fix partial z/s clears with sysmem.

We have to set 8c01 to say "leave these channels alone" when
clearing/storing just Z or S of z24s8.  Fixes the bypass path for
KHR-GLES3.packed_depth_stencil.verify_read_pixels.depth24_stencil8.

Cc: mesa-stable
Fixes: #5592
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13659>

features: add dynamic render for lavapipe

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13627>

lavapipe: VK_KHR_dynamic_rendering

this is a conformant implementation

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13627>

lavapipe: fix cmd queuing for dynamic render

this sucks, but it should be sorted quickly in #5440

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13627>

vk: update headers for 1.2.197

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13627>

freedreno/ir3: Use immediate for flat.b's src1

According to Jonathan Marek:

  Only one immediate can be decoded in a cat2 instruction (if both srcs
  are immediates, they will use the value of the either the first or
  second one, I don't remember which) - using 2 immediates in a cat2
  instruction is only "correct" if they are both equal.

  The (i,j) in the second src of flat.b is not unused, but behaves as 0
  for any (small) integer because it is a float src. The hack I
  suggested is to set the second src equal to (immediate) first src,
  which seems to work.

This allows us to remove a couple of mov instructions or a bit of extra
constfile usage.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13558>

freedreno/ir3: Use flat.b to load flat varyings on a6xx

The flat.b/bary.f cat2 instruction should be faster than an ldlv cat6
instruction, even with a couple of additional moves (which will be
removed in the next patch).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13558>

freedreno/ir3: Add infrastructure for flat.b

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13558>

ir3: Add support for (dis)assembling flat.b

flat.b is a variant of the bary.f instruction that does not perform
interpolation of the varying input.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13558>

zink: always use lazy (non-push) updating for fbfetch descriptors

fbfetch descriptors are uncacheable due to having mixed descriptor types
in the same set, so this needs to always use lazy updating to avoid
exploding the cache and crashing

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13654>

zink: set fbfetch state on lazy batch data when enabling it

this avoids creating new descriptor pools on every update

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13654>

anv: disable debug logging spam

nobody wants to see this by default, so disable it until #5404
is resolved to make it more manageable

Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13628>

mesa/st: lower psiz for shader precompile

if the driver is requesting that psiz always be set, precompile the first
variant with psiz since that's most likely to be what's eventually used

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13588>

mesa/st: rework psiz lowering

the context flag is immutable, and it's set by drivers that always
require the shader to export a psiz value for rasterization

"always" is all cases except GL_VERTEX_PROGRAM_POINT_SIZE being
enabled, in which case it should be assumed that the shader already
writes it, and no changes should be made

thus the shader key value can be changed to reflect that, when set,
it requires the shader to export a psiz value if it doesn't already

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13588>

docs: update calendar for 21.3.0-rc4

Add another release candidate as we're not ready for 21.3.0 final just yet.

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13667>

st/mesa: don't update vertex elements when GL doesn't change them

We rely on mesa/main to tell us whether to update vertex elements.
This decreases overhead for obvious reasons.

The select/feedback code path doesn't use this, which is why you see
unconditional "ALL" in a few codepaths.

This sequence of GL calls doesn't update vertex elements if only
the pointer and stride vary:
    glVertexPointer()
    glDrawElements()
    glVertexPointer()
    glDrawElements()
    glVertexPointer()
    glDrawElements()

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13512>

cso: add missing parameters into cso_set_vertex_buffers

they will be used later

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13512>

mesa: add NewVertexBuffers/NewVertexElements flags to indicate state changes

This splits the per-VAO NewArrays flag into NewVertexBuffers and
NewVertexElements, and adds a global NewVertexElements flag to be used
by drivers.

This allows gallium to skip updating vertex elements when only vertex
buffers need to be updated.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13512>

mesa: change gl_vertex_array_object::NewArrays to bool

the individual bits are never used

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13512>

st/mesa: use POPCNT in st_update_array if the CPU supports it

The st_update_array overhead decreases from 8.28% to 7.67% for a viewperf
subtest.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13512>

st/mesa: change st_atom_array.c to cpp

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13512>

util: add a util_bitcount variant that selects POPCNT through C++ template arg

Moved from radeonsi. st/mesa will use it.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13512>

r300: stub derivatives on r300 and r400 hardware

There are three approaches for problem:
- use dummy shader (current solution)
- use software rendering
- stub

First option always gonna give bad results, second one
gonna be slideshow (r300/r400 at best are running with
old 2 cores cpu), third one can sometimes give graphical
gliches.

IMHO third one is least annoying option.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13642>

freedreno: Fix layered rendering to just Z/S and not color.

We would try to take the gmem path which can't do layered rendering.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13653>

build: add sha1_h to llvmpipe build

cc: mesa-stable

fixes #5588

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13658>

.mailmap: Simplify my name

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13651>

zink: add queue locking

sparse binds have to be processed synchronously with cmdbuf recording to
avoid resource object desync in the vk driver, which means they have to be
done in the driver thread instead of the flush thread. this necessitates
adding locking for the queue since there is now a case when submissions occur
in a different thread

fixes illegal multithread usage in KHR-GL46.CommonBugs.CommonBug_SparseBuffersWithCopyOps

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13597>

zink: set PIPE_CAP_VERTEX_ATTRIB_ELEMENT_ALIGNED_ONLY

fixes #5557

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13556>

gallium: add PIPE_CAP_VERTEX_ATTRIB_ELEMENT_ALIGNED_ONLY

vulkan requires that vertex attribute access be aligned to the size of
a component for the attribute, but GL has no such requirements

the existing alignment caps are unnecessarily restrictive for applying
this limitation, so this cap now pre-calculates the masks for elements
and vertex buffers in vbuf to enable rewriting misaligned buffers

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13556>

freedreno/a6xx: Use the fdl buffer view setup for img/ssbo descriptors.

The single-plane descriptor emit helper doesn't strictly need the UBWC
reloc, since imageBuffer can't be UBWC, but it means the function is ready
to be used for non-buffer image descriptors later.

no-hw drawoverhead 1-imageBuffer change throughput 1.95457% +/- 1.44325%
(n=127).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13635>

freedreno/fdl6: Add support for texture swizzles of A/L/I/LA/RGBx.

To convert freedreno over, we need to support these formats where we remap
R or RG formats to GL compat ones, or RGBA to RGBx.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13635>

turnip: Remove buffer-view cross-check code.

Now that I've tested storage.*buffer, I'm confident I've moved the buffer
views correctly.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13635>

turnip: Use the new shared buffer-view descriptor creation function.

This cross-checks that our descriptors match as I move the code.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13635>

freedreno/fdl6: Add an interface for setting up buffer descriptors.

Buffers don't need all the layout stuff that image views do, so it's
easier to have a separate interface for generating them.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13635>

freedreno/a6xx: Emit a null descriptor for unoccupied IBO slots.

Fixes a crash in some desktop GL testcases in piglit.

Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13635>

freedreno: Fix gmem invalidating the depth or stencil of packed d/s.

The gmem store stores both depth and stencil for z24s8.  So, if we're
doing a write (clear or draw) to one or the other of the channels, we need
the other one restored as well.

Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13649>

intel/compiler: Use gl_shader_stage_uses_workgroup() helpers

Instead of checking for MESA_SHADER_COMPUTE (and KERNEL).  Where
appropriate, also use gl_shader_stage_is_compute().

This allows most of the workgroup-related lowering to be applied to
Task and Mesh shaders.  These will be added later and "inherit" from
cs_prog_data structure.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13629>

intel/compiler: Make brw_nir_populate_wm_prog_data() static

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13629>

anv: Include viewport size in scissor rectangle

Prevent drawing outside the viewport when viewport size
is smaller than framebuffer size.

v2: (Jason Ekstrand)
- re-emit scissor on viewport change
- do the same calculations for other platforms

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5515
Signed-off-by: Vadym Shovkoplias <vadym.shovkoplias@globallogic.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13594>

anv: Process FS last when compiling graphics pipeline

Enum values for MESA_SHADER_TASK and MESA_SHADER_MESH are larger than
MESA_SHADER_FRAGMENT, so can't rely on the them for ordering anymore.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13637>

anv: Make shaders array in anv_graphics_pipeline fit Task/Mesh

We could decouple the locations in the array from the gl_shader_stage
enum values, but for now this is convenient.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13637>

anv: Get rid of "may be used initialized" warning in anv_QueueSubmit2KHR

The code is correct, but compiler can't see it.  Initialize the value
to NULL and assert on it if the function succeeds.  It both helps the
compiler and make the code slightly more robust.

```
../src/intel/vulkan/anv_queue.c: In function ‘anv_QueueSubmit2KHR’:
../src/intel/vulkan/anv_queue.c:932:16: warning: ‘impl’ may be used uninitialized in this function [-Wmaybe-uninitialized]
  932 |       result = anv_queue_submit_add_timeline_wait(queue, submit,
      |                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  933 |                                                   &impl->timeline,
      |                                                   ~~~~~~~~~~~~~~~~
  934 |                                                   value);
      |                                                   ~~~~~~
../src/intel/vulkan/anv_queue.c:899:31: note: ‘impl’ was declared here
  899 |    struct anv_semaphore_impl *impl;
      |                               ^~~~
```

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13638>

freedreno/computerator: Support A660 gpu

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13640>

tu: fix rast state allocation size on a6xx gen4

A few regs were added without changing the size of draw state.

Fixes: 4e05338d99abbf2858a0d8444ffc53028fe23051 ("turnip: Rast updates for a6xx gen4")

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13644>

zink: reject all storage multisampling if the feature is unsupported

this also enables removing a stupid conditional

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13631>

zink: add SpvCapabilityStorageImageMultisample for multisampled storage images

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13631>

broadcom/compiler: check that sig packing is valid when pipelining ldvary

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13641>

glsl/nir: mark samplers inside a block as bindless

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13416>

mesa: don't reset SamplersValidated if nothing changed

This could prevent error detection, if a uniform change sets
SamplersValidated to true without calling _mesa_update_shader_textures_used.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13416>

radv: optimize subpass barrier flushes for imageless framebuffers

The driver should always know the attachments at this point. This
should reduce the number of L2 cache flushes for imageless framebuffers.

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/13291>

intel/compiler: Initialize SIMDSelectionTest member error.

Fix defect reported by Coverity Scan.

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

Fixes: 7558340ebb6 ("intel/compiler: Add helpers to select SIMD for compute shaders")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13608>

intel/compiler: Change selected_simd return type to int.

brw_simd_select return type is int.

Fix defect reported by Coverity Scan.

Unsigned compared against 0 (NO_EFFECT)
unsigned_compare: This less-than-zero comparison of an unsigned value is never true. selected_simd < 0U.

Fixes: 7dda0cf2b8e ("intel/compiler: Use SIMD selection helpers for CS")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13606>

zink: stop double printing validation messages

VVL already prints its messages using configurable settings. there's no
reason for zink to unconditionally repeat them immediately after

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13633>

ci: Uprev VK-GL-CTS to 1.2.7.2, and pull in piglit while I'm here.

The VK-GL-CTS fixes some issues for freedreno, and almost all of LVP's
xfails.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13622>

microsoft/compiler: Use textures for SRVs

After running the (renamed) dxil_nir_split_typed_samplers pass, the
shader will have either:
* Textures, which map to D3D SRVs
* Bare samplers, which map to D3D bare samplers
* Images, which map to D3D UAVs

There shouldn't be any remaining samplers with type information

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13390>

util/hash_table: Clear special 0/1 entries for u64 hash table too

Fixes: e532a47f ("util/hash_table: do not leak u64 struct key")
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13390>

docs: update radv extensions in features.txt

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13626>

broadcom/compiler: make opt passes set current block

Typically, optimization passes go through all the blocks in a shader
and make adjustments on the fly, so we always want them to update
the current block or the current block pointer will become outdated.

Also, we don't need to keep track of the previous current block
pointer to restore it, since optimization passes run after we have
completed conversion to VIR, and therefore, anything that comes after
that should always set the current block before emitting code.

Fixes debug assert crashes when running shader-db:
vir.c:1888: try_opt_ldunif: Assertion `found || &c->cur_block->instructions == c->cursor.link' failed

Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13625>

freedreno/ci: Test Turnip on Adreno 618

Collabora has added 7 new lazor Chromebooks which have Adreno 618 GPUs.

Run half of the test suite (plus variants) in pre-merge.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13441>

ci: Let manual LAVA jobs have a longer timeout than others

So far only LAVA jobs make use of it, but I guess baremetal could be
extended to have these timeouts as well.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13441>

ci: Add support for lazor Chromebooks

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13441>

virgl: Add driconf tweak to force-enable reading back R8_SRGB textures

In the menu of CS:GO R8_SRGB textures are uploaded and read back, and
since R8_SRGB can't be read back on GLES, because it is not a rendertarget
format and glGetTexImage and siblings don't exists, we can't default to
enabling reading back this format. This leads to an emulation of the
glGetTexImage calls issued by CS:GO, and this slows down the menus a lot
(below 1 fps on Intel XE hosts).

So add this driconf tweak and enable it for CS:GO to work around the issue.
It can be done safely, because in this case we actually can use the data
that is stored on the host in the backing IOV.

This tweak lets the CS:GO menu run at around 60 FPS when run with virgl
on a Intel XE host when it would run with less than 1 FPS without the tweak.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: John Bates <jbates@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13572>

iris: fix off-by-one error when clearing stale syncobjs

This shouldn't fix any real world bugs, except it will now clear more
stale syncobjs than it was previously doing, and actually do what the
comment says it does.

I could not find a real workload where this change would be relevant,
although I didn't try too much. I wrote my own little egl program to
test this.

I spotted this while reading the code when investigating a Piglit
failure [0]. It turns out this part the code was not relvant for the
failure.

[0]: ext_external_objects-vk-image-display-overwrite

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13536>

radv: only enable VK_EXT_display_control for vrcompositor (SteamVR)

One CTS test is still failing and the fix isn't upstream yet.

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/13623>

v3dv: Implement VK_KHR_create_renderpass2

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13575>

panfrost/ci: update piglit tests expectations on G52

Because we aren't running this test on pre-merge, the expectations have
gone out of sync with the code. Sync them again.

Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13089>

ci/freedreno: Add maxcpus=2 to the kernel cmdline on a530

It seems that the cpufreq support in a530 is not rock solid yet, as the
device intermittently reboots during the boot process.

Passing `maxcpus=2` to the kernel avoids this issue.

Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Acked-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13089>

iris/ci: Fix traces for amly and deqp list for whl

These are manual jobs, so the expected results got out of date.

Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13089>

ci: Update linux kernel to v5.15

* Update Kconfig for x86_64 and ARM64. Follow the dependency tree of the
  kernel modules to make sure that the intended configurations are being
  set. Check scripts/merge_config.sh output as well to see if there is
  a requested Kconfig not being considered.

For a630 devices:
* Use kernel version with a6xx workaround for frequency scaling
* Enable CONFIG_QCOM_LMH targeting a630 slowness on new kernel

---- Out of tree patches used ----

For a360 device:
* Revert a commit which remove slpi_region from msm8996:
  8b0031f8bda2 ("Revert "arm64: msm8996: fix memory region overlap"")

Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13089>

xmlconfig_test: add unit test for executable_regexp

Reviewed-by: Emma Anholt <emma@anholt.net>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13304>

drirc: add Mari application workaround

To allow using non-constant sampler array index. This prevent
Mari render fail for any scene.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13304>

driconf: add executable_regexp application attribute

For matching executable with variable names like Mari which
append version string to its executable like: Mari4.5v2, Mari4.7v4.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13304>

zink: Remove duplicate variable unsized.

Fix defect reported by Coverity Scan.

Evaluation order violation (EVALUATION_ORDER)
write_write_typo: In unsized = unsized =
glsl_array_type(glsl_uintN_t_type(bit_size), 0U, bit_size / 8U),
unsized is written twice with the same value.

Fixes: f79a25653b9 ("zink: move all shader bo/sharedmem access to compiler passes")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13607>

vulkan/wsi: set correct bits for host allocations/exports for images.

Lavapipe was hitting asserts in this area due to incorrect bits being
specified.

Set the handle type depending on the sw flag, and set a correct handle
type for the memory host ptrs.

v2: add image export struct to image creation (Jason)

Fixes: 895d3399f7fb ("lavapipe: add support for KHR_external_memory_fd")
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13615>

radv: Disable coherent L2 optimization on cards with noncoherent L2.

With high likelihood we are forgetting to set the noncoherent bits
somewhere but I don't have the HW to debug. To avoid user pain
disable this optimization on these GPUs.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5505
Fixes: fd8210f27ed ("radv: Try to do a better job of dealing with L2 coherent images.")
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13620>

frontends/va: Return error in vaRenderPicture if decoder is NULL

This fixes a crash if a data slice is submitted before the decoder
is initialized. A well-behaved application shouldn't encounter this
but returning an error is still better than crashing the entire
process and the rest of the code is similarly defensive.

Signed-off-by: Lorenz Brun <lorenz@brun.one>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11792>

intel/fs: Fix a cmod prop bug when cmod is set to inst that doesn't support it

Fixes dEQP-VK.reconvergence.*nesting* tests.

There are cases when cmod is set to an instruction that cannot have
conditional modifier. E.g. following:

 find_live_channel(32) vgrf166:UD,  NoMask
 cmp.z.f0.0(32) null:D, vgrf166+0.0<0>:D, 0d

is optimized to:

 find_live_channel.z.f0.0(32) vgrf166:UD,  NoMask

v2:
 - Add unit test to check cmod is not set to 'find_live_channel' (Matt Turner)
 - Update flag_subreg when conditonal_mod is updated (Ian Romanick)

Signed-off-by: Vadym Shovkoplias <vadym.shovkoplias@globallogic.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5431
Fixes: 32b7ba66b01 ("intel/compiler: fix cmod propagation optimisations")
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13268>