vulkan/util: generate define for a selected few enums

One of the unfortunate effect of Vulkan starting to use 64bit bitmasks
is that they can no longer be defined using enums because C doesn't
guarantees that enum values will be 64bits.

Vulkan therefore started using those patterns :

  static const VkAccessFlags2KHR VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT_KHR = 0x00000001;

This has the effect that we can not longer use those values in
switch/case statements.

This change introduces defines so that we can keep doing this. For now
only VkAccessFlags2KHR/VkPipelineStageFlags2KHR are allowed to be
redefined this way, this list could be changed later (or all bitmask
could be processed this way).

v2: Generate hexadecimal numbers (Jason)

Signed-off-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/9045>

vulkan: remove unused VkCommand

Signed-off-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/9045>

vulkan: handle new VK_KHR_synchronization2 image layouts

Signed-off-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/9045>

vulkan: put generated defines into their own header

v2: Put vk_enum_defines.h in the list of generated files

Signed-off-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/9045>

wsi/x11: cleanup properly after mit shm paths are used.

This path was being skipped on sw + has_mit_shm, when it
shouldn't be.

Fixes: a069b4e9b9a1 ("vulkan/wsi/sw: add mit-shm support for pixmap allocation")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13217>

docs: mark off missing lavapipe exts

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13232>

turnip: Free disk cache on pdev init failure.

Noticed while debugging test failure under valgrind (the disk cache
doesn't come from the vulkan allocator, so we could leak it and not fail
the test).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13200>

turnip: Fix allocation failure handling around device->name.

Fixes regressions in dEQP-VK.api.device_init.create_instance_device_intentional_alloc_fail

Fixes: 5116388e0ba9 ("turnip: Expose a device name similar to the blob.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13200>

turnip: Plug the vendor/device ID into the pipeline cache fields, too.

Fixes a regression in dEQP-VK.pipeline.cache.misc_tests.cache_header_test.

Fixes: 7e471541e0d1 ("turnip: Match the blob's format for vendorID and deviceID.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13200>

ir3: remove obsolete assert for intrinsic_store_output in tess

We do support non-zero nir_intrinsic_component for
nir_intrinsic_store_output in tess shaders.

Fixes vkd3d test "test_hull_shader_fork_phase_dxbc"

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13196>

anv: allocate zeroed device object

When enabling a new feature we made the mistake of initializing some fields
of the device object conditionally, which leads to crashes later. Initializing
those fields would be a trivial fix, but it's probably better to just zero
everything at allocation time and prevent any future screwups. Device objects
are allocated rarely enough for this additional memset to not matter for
performance.

Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
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/13221>

genxml: Drop bit 27 from RENDER_SURFACE_STATE::Surface Format

Bit 27 is the "ASTC Format" bit in the PRMs but we just extended the
Surface Format field by one bit and made sure all the ASTC formats have
that bit set.  Since Gfx12.5 doesn't support ASTC, we can drop that bit
from the field and this will cause GenXML packing functions to assert if
it's ever set.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13206>

intel/isl: ASTC support was removed on Gfx12.5

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13206>

anv: Ask ISL about ASTC support

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13206>

intel/isl: Stop claiming ASTC works on Cherry View

We do, however, leave a nice tombstone comment in case anyone comes
looking.  Given the age and scarcity of Cherry View hardware and ASTC
apps that run on desktop Linux, it's unlikely we'll ever bother to
implement it.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13206>

intel/isl: Simplify isl_format_supports_filtering

For compressed formats, filtering and sampling has always gone together.
This lets us avoid duplicating all those nasty special cases between the
two functions.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13206>

build: unify vulkan cpp platform args

these were duplicated all over the place, and it's annoying to have to keep
duplicating them any time a new component includes the vulkan header

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13141>

lavapipe: remove display extension support

lavapipe doesn't actually support these

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Acked-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13141>

lavapipe: enable KHR_external_memory_fd

Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Heinrich Fink <hfink@snap.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12345>

llvmpipe: enable EXT_memory_object(_fd)

Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Heinrich Fink <hfink@snap.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12345>

lavapipe: add support for KHR_external_memory_fd

Support creating exportable memory. Use memfd file
descriptors and import/export them as opaque fd handles.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Heinrich Fink <hfink@snap.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12345>

llvmpipe: add support for EXT_memory_object(_fd)

Enable the import of memory via opaque fd handles, which
are based upon memory-fds. The extension is necessary for sharing
images and buffers from Vulkan.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Heinrich Fink <hfink@snap.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12345>

gallium: add utility and interface for memory fd allocations

Add utility functions to allocate aligned memory backed by
mem_fd objects. Add interface to Gallium for same allocation.
It will be used in later commits for external memory support
in Vulkan/OpenGL.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Heinrich Fink <hfink@snap.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12345>

ir3: Use source in ir3_output_conv_src_type()

This was incorrectly converted when splitting the regs array. Noticed by
inspection.

Fixes: d3e08327cf8 ("ir3/core: Switch to srcs/dsts arrays")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13220>

meson_options: Bump max value of platform-sdk-version to 31

During building Android-12, the following error appears:
meson.build:21:0: ERROR: New value 31 is more than maximum value 30.

Signed-off-by: Roman Stratiienko <r.stratiienko@gmail.com>
Acked-by: Emil Velikov <emil.l.velikov@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13216>

turnip: fix vbs emission when there are holes in bindings

Otherwise we read garbage for bindings with value above
vertexBindingDescriptionCount.

Fixes vkd3d test "test_append_aligned_element"

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13195>

v3dv: use NULL for vk_error on initialization failures

This commit fixes two issues:

  * On CreateInstance, we are freeing the instance, and then trying to
    use it when calling vk_error. This could be problematic, so let's
    just use NULL.

  * On CreateDevice, we are getting a unsupported feature error, and
    then trying to call vk_error using the instance. That's is not
    really a instance error, and will assert when the ongoing common
    vk_error lands mesa. Let's use NULL instead, as the object it
    applies, the device, was not created.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13219>

device_select: close dri3 fd after using it.

This can leak and causes crashes in some CTS test groups
dEQP-VK.wsi.xcb.incremental_present*

Fixes: 9bc5b2d169d3 ("vulkan: add initial device selection layer. (v6.1)")
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13215>

lvp/fence: quick fix to previous commit.

This fixes last of xcb cts issues.

Fixes: 8a294b6f975e1 ("lavapipe: Fix vkWaitForFences for initially-signalled fences")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13212>

lavapipe: Fix vkWaitForFences for initially-signalled fences

Fences with VK_FENCE_CREATE_SIGNALED_BIT are created with
signalled=true and timeline=0, waiting on them without
submitting first returned VK_TIMEOUT instead of VK_SUCCESS.

Signed-off-by: Pavel Asyutchenko <sventeam@yandex.ru>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13128>

lavapipe: EXT_4444_formats support

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12001>

llvmpipe: fix 4-bit output scaling.

This is overkill, but hey 4-bits per channel is hardly something to
care about.

(Suggestions welcome for a better version).

Fixes:
dEQP-GLES2.functional.fbo.render.*rgba4*

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12001>

virgl: Add support for NIR shaders when VIRGL_DEBUG=nir.

This will let me incrementally fix nir-to-tgsi against virgl without
having to carry around the whole "remove TGSI from mesa/st" MR.

Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Force the TXQ LOD argument to be scalar.

Otherwise, older virglrenderer fails all the texturesize tests.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Add a workaround for virgl UBO array dynamic indexing.

virgl makes one array of UBOs starting from the first non-CB0 UBO used,
and does dynamic indexing off of that.  It requires that the dynamic
indexing be CONST[ADDR[0]+base], rather than having the base be loaded in
addr0.

If we had a nir_intrinsic_base() on load_ubo, this would be easy.  As we
don't, emit a subtract at address deref time.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Sort FS output declarations to avoid virglrenderer bugs.

The TGSI debug output is a lot more readable if it's in location order,
anyway.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir-to-tgsi: Avoid emitting TXL just for lod 0 on non-vertex shaders.

Prompted by comparing virgl fails and finding that it has issues with
immediate args to TXL/TXB, at least.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Turn GS PRIMID into an input instead of a sysval.

While TGSI can represent it either way, virgl and r600 at least demand an
input.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Add support for nir_intrinsic_load_barycentric_at_sample.

It doesn't have to be a constant sample, so we need to store it at load
time and use the load's dest at interpolate_at time.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Add support for load_barycentric_sample.

This is used for var->data.sample inputs, which are already declared to be
TGSI_INTERPOLATE_LOC_SAMPLE, so we can just use the interpolated inputs.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Add support for declaring image arrays.

Required for virgl.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

gallium/ureg: Sort the input decls, too.

Just like outputs, virglrenderer needs its inputs sorted.  Should be
harmless for other TGSI producers, and makes the declarations more
readable.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Add support for load_output/load_per_vertex_output.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

nir_to_tgsi: Include txf_ms's sample index.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

mesa/st: Don't bump locations of patch vars for !PIPE_CAP_TEXCOORD.

There's no need to reserve the bottom 9 VARYING_SLOT_PATCH*, since
VARYING_SLOT_TEXCOORD won't be mapped there.  This helps us match up with
nir_to_tgsi, which wasn't shifting down by 9 for patch.

Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12800>

lavapipe: inherit from vk_image

simple and easy since we don't use much of this anyway

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13146>

llvmpipe/cs: change submission pattern for threadpool

Recent ncnn benchmarks showed a slowdown, and this change seemed
more likely.

The batching into threads for the main workloads is fine, however
the remainder stuff doesn't get spread out and can bottleneck in
one thread.

Switch to a model where the initial work is batched, but the
remainder is iterated over one by one.

Brings ncnn benchmarks back in line with previously.

Fixes: 69109e0b1935 ("llvmpipe/cs: rework thread pool for avoid mtx locking")
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13210>

anv: enable VK_KHR_maintenance4

v2 (Jason Ekstrand):
 - Get maxBufferSize from ISL.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13199>

intel/isl: Add a max_buffer_size limit to isl_device

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13199>

anv: implement vkGetDeviceImageSparseMemoryRequirementsKHR

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13199>

anv: implement vkGetDeviceImageMemoryRequirementsKHR

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13199>

anv: move VkImage object allocation to anv_CreateImage

v2 (Jason Ekstrand):
 - Switch the order of arguments to be device, image, other stuff

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13199>

anv: Add an anv_image_get_memory_requirements helper

This is similar to a patch from Lionel except works in terms of aspects
rather than bindings.  This makes it easy to use from the Android code.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13199>

anv: remove unused function

Fixes: 49908c602ffd ("anv/android: Rework our handling of AHardwareBuffer imports")
Signed-off-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/13199>

anv: implement vkGetDeviceBufferMemoryRequirementsKHR

Signed-off-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/13199>

anv: move GetBufferMemoryRequirement with other buffer functions

Signed-off-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/13199>

vulkan: Update the XML and headers to 1.2.195

Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13199>

nir/algebraic: Small optimizations for SpvOpFOrdNotEqual and SpvOpFUnordEqual

No shader-db changes on any Intel platform.

Fossil-db results:

All Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 144380118 -> 143692823 (-0.5%)
SENDs in all programs: 6920822 -> 6920822 (+0.0%)
Loops in all programs: 38299 -> 38299 (+0.0%)
Cycles in all programs: 8434782176 -> 8423078994 (-0.1%)
Spills in all programs: 206830 -> 204469 (-1.1%)
Fills in all programs: 318737 -> 313660 (-1.6%)

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12320>

spirv: Generate shorter code for SpvOpFUnord comparisons

No shader-db or fossil-db changes on any Intel platform.

v2: Keep the flt <-> fge switcharoo local to the SpvOpFUnordLessThan,
etc. handling.  Add a comment explaining why the suboptimal
SpvOpFUnordEqual implementation is used here.  Suggested by Caio.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12320>

spirv: SpvOpFUnordNotEqual doesn't need special treatment

The NIR fneu opcode already matches the "unordered not equal" semantics
of the SPIR-V opcode.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12320>

spirv: Minor cleanup in SpvOpFOrdNotEqual

v2: Add a comment explaining why the suboptimal SpvOpFOrdNotEqual
implementation is still used here.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12320>

spirv: Silence unused parameter warnings in vtn_alu.c

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12320>

ci: updates

fails are from #4571

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12831>

zink: support 16bit rgbx formats

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12831>

pan/bi: Optimize abs(derivative)

We implement fine derivatives as:

broadcast(x, (lane & ~1) + 1) - broadcast(x, lane & ~1)

Most of the complexity is to get the right sign. If we can ignore the
sign, we can generate the simpler code:

broadcast(x, lane ^ 1) - lane

This is a particular win on v7+ where the broadcast instruction (CLPER)
can do `lane ^ value` for free. However, even on v6 where we lower to an
explicit XOR instruction, it's still a win.

The limiting case is fwidth. The fragment shader

   gl_FragColor = fwidth(vec4_varying);

has the following results on v6, v7, and v9:

G72 (-26% instructions, -43% cycles):
38 inst, 30 tuples, 5 clauses, 1.166667 cycles, 1.166667 arith, 28 quadwords
28 inst, 19 tuples, 4 clauses, 0.666667 cycles, 0.666667 arith, 19 quadwords

G76 (-37% instructions, -54% cycles):
38 inst, 30 tuples, 5 clauses, 1.166667 cycles, 1.166667 arith, 28 quadwords
24 inst, 16 tuples, 4 clauses, 0.541667 cycles, 0.541667 arith, 18 quadwords

G78 (-40% instructions, -56% cycles):
40 inst, 1.125000 cycles, 0.250000 fma, 0.109375 cvt, 1.125000 sfu, 20 quadwords
24 inst, 0.500000 cycles, 0.250000 fma, 0.015625 cvt, 0.500000 sfu, 12 quadwords

shader-db tells a similar story -- most shaders are unaffected, but a
shader that uses fwidth has a 20% reduction in cycle count:

instructions helped:   shaders/tesseract/488.shader_test MESA_SHADER_FRAGMENT: 264 -> 262 (-0.76%)
instructions helped:   shaders/chromeos/109-1.shader_test MESA_SHADER_FRAGMENT: 36 -> 28 (-22.22%)
tuples helped:   shaders/chromeos/109-1.shader_test MESA_SHADER_FRAGMENT: 27 -> 22 (-18.52%)
tuples HURT:   shaders/tesseract/488.shader_test MESA_SHADER_FRAGMENT: 211 -> 212 (0.47%)
clauses HURT:   shaders/tesseract/488.shader_test MESA_SHADER_FRAGMENT: 32 -> 33 (3.12%)
cycles helped:   shaders/chromeos/109-1.shader_test MESA_SHADER_FRAGMENT: 1 -> 0.79 (-20.83%)
arith helped:   shaders/chromeos/109-1.shader_test MESA_SHADER_FRAGMENT: 1 -> 0.79 (-20.83%)
quadwords helped:   shaders/chromeos/109-1.shader_test MESA_SHADER_FRAGMENT: 31 -> 28 (-9.68%)
quadwords HURT:   shaders/tesseract/488.shader_test MESA_SHADER_FRAGMENT: 176 -> 178 (1.14%)

total instructions in shared programs: 148370 -> 148360 (<.01%)
instructions in affected programs: 300 -> 290 (-3.33%)
helped: 2
HURT: 0

total tuples in shared programs: 124188 -> 124184 (<.01%)
tuples in affected programs: 238 -> 234 (-1.68%)
helped: 1
HURT: 1
helped stats (abs) min: 5.0 max: 5.0 x̄: 5.00 x̃: 5
helped stats (rel) min: 18.52% max: 18.52% x̄: 18.52% x̃: 18.52%
HURT stats (abs)   min: 1.0 max: 1.0 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 0.47% max: 0.47% x̄: 0.47% x̃: 0.47%

total clauses in shared programs: 25692 -> 25693 (<.01%)
clauses in affected programs: 32 -> 33 (3.12%)
helped: 0
HURT: 1

total cycles in shared programs: 12132.04 -> 12131.83 (<.01%)
cycles in affected programs: 1 -> 0.79 (-20.83%)
helped: 1
HURT: 0

total arith in shared programs: 4623.75 -> 4623.54 (<.01%)
arith in affected programs: 1 -> 0.79 (-20.83%)
helped: 1
HURT: 0

total quadwords in shared programs: 110386 -> 110385 (<.01%)
quadwords in affected programs: 207 -> 206 (-0.48%)
helped: 1
HURT: 1
helped stats (abs) min: 3.0 max: 3.0 x̄: 3.00 x̃: 3
helped stats (rel) min: 9.68% max: 9.68% x̄: 9.68% x̃: 9.68%
HURT stats (abs)   min: 2.0 max: 2.0 x̄: 2.00 x̃: 2
HURT stats (rel)   min: 1.14% max: 1.14% x̄: 1.14% x̃: 1.14%

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12332>

nir: Add Mali-specific derivative opcodes

Add derivative opcodes fddx_must_abs_mali/fddy_must_abs_mali satisfying:

   fabs(fdd*_must_abs_mali(v)) = fabs(fdd*(v))

The sign of their result is undefined.

On Bifrost and Valhall, these unsigned derivatives can be implemented
more efficiently than the correctly-signed counterparts, since the sign
fixup requires extra ALU instructions. On backends where this is the
case, it is useful to optimize fabs(fdd*(v)) to
fabs(fdd*_must_abs_mali(v)). This pattern comes up with the GLSL builtin
`fwidth`.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12332>

ac/surface: Use 64 && 128 for GFX10_3 on non-modifier path

DCC_IND_BLK is not hooked up for this to work in the kernel in any released version, and it's unsafe to do so even if it was because it doesn't check the modifiers.

There's no reason to change the legacy non-modifier path to be more performant at the expense of breaking backwards compatibility with older versions of Mesa.

Fixes: 0f6251b3 ("ac/surface: use DCC compatible with image stores for < 4K resolutions")
Closes: #5422

Signed-off-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13122>

mesa: use simple_mtx_t for ShaderIncludeMutex

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13152>

mesa: use simple_mtx_t for TexMutex

change mtx_recursive -> mtx_plain, there's no recursive locking

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13152>

gallium/pb_slab: use simple_mtx_t

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13152>

gallium/pb_cache: use simple_mtx_t

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13152>

util/queue: use simple_mtx_t for finish_lock

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13152>

util/slab: use simple_mtx_t

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13152>

mesa: add missing unlock_texture into generate_texture_mipmap

Fixes: 5a39938b0004de36ac55 "mesa: Throw an error for compressed glGenerateMipmap on GLES2 contexts."

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13152>

nir_to_dxil: Add tagging raw SRVs in shader flags

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13157>

gallivm/format: clamp SINT conversion rather than truncate.

KHR-GL32.packed_pixels.pbo_rectangle.r16i on zink on lavapipe
ends up using a pbo that does an SINT image write. This was producing
truncated rather than clamped values.

Fix the calculations for 8/16-bit signed ints to clamp not truncate.

Fixes: 13e5f331db77 ("gallivm/nir: fix image store conversions")

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13187>

i915g: Check for the scanout-layout conditions before setting level info.

Fixes leaks (release) or assertion failures (debug) on allocating small
scanout resources, when falling through to the non-scanout-specific layout
code, which became more common as of ad50b47a14e9 ("gbm: assume
USE_SCANOUT in create_with_modifiers").

Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13202>

panfrost: Fix PAN_MESA_DEBUG=sync with INTEL_blackhole_render

We need to ignore the "sync" in this case, or we'll crash with
"incomplete job" since we never submitted the work. Fixes the Piglit
intel_blackhole-draw_gles2 when run in CI.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13070>

panfrost: Don't set CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER

We already set HALF_INTEGER, which is what the compiler actually does.
If we also set PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER, we get
incorrect lowering. Only set the CAP we respect.

On Bifrost, this convention is arbitrary. We should consider moving the
Bifrost lowering into NIR to optimize this better...

Fixes Piglit glsl-arb-fragment-coord-conventions.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13070>

panfrost: Introduce PAN_MAX_VARYINGS define

The number of varying records we need to reserve in the worst case is
greater than the number of source-level varyings we advertise
(gl_Position, gl_PointSize...)

We advertise MAX_VARYINGS source level varyings, which means anywhere we
manipulate varyings we need up to (MAX_VARYINGS + max non-source level
varyings) records. Add a PAN_MAX_VARYINGS define for this and use it
throughout.

Fixes a buffer overflow in Piglit glsl-max-varyings, which now passes
instead of crashes.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13070>

panfrost: Fix off-by-one in varying count assert

We want to assert that the number of varyings (the count) is at most the
the maximum count. This is <=, not <, with the assertion previously
failing for exactly the maximum.

Fixes: 2c2cf0ecfe6 ("panfrost: Streamline varying linking code")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13070>

panfrost: Move special_varying to compiler definitions

The number of special varyings on midgard can influence how much space
we need to allocate for varyings in the compiler ABI. Move the enum so
we can get access it.

No functional change. This is cc stable purely so the following patches
can be backported.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13070>

llvmpipe: fix crash when doing FB fetch + gl_FragDepth write in one shader

Reproducible by piglit test from this MR:
https://gitlab.freedesktop.org/mesa/piglit/-/merge_requests/576

Signed-off-by: Pavel Asyutchenko <sventeam@yandex.ru>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12705>

turnip: Replace our format table with fd6_format_table.

Now we have a common format table with just a little bit of overriding in
turnip, which I hope we can decrease over time.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

turnip: Do format lookups from the fd6 format table and cross-check.

Intermediate step to make sure things stay the same as I transition us to
sharing the format table.

The YUV special casing is pretty not the prettiest, but I'm hoping to
continue working on planar RGB formats as pipe formats separately.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

turnip: Switch tu_format internals to using pipe_format more.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

turnip: Give D32_SFLOAT_S8_UINT a native format.

We won't actually use the native format here, as when sampling our image
we we just see the format for either the D or S aspect.  This syncs us up
with freedreno's table, though.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Enable UBWC for RGBA5551 (and 1555) textures.

Turnip has it enabled, and it seems to be happy in the CTS.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Add support for A/XRGB1555 formats.

This switches the format we use by default for GLES's rgb5_a1, and means
that our formats supported more closely matches turnip.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Rewrite the format table format/swap helpers.

We need to take a tiling mode to in order to support 1555 like turnip
does, and I like this naming better.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Make the format table const.

This data never changes.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Use fd6_pipe2tex() for the 2D src format.

The differences between the two are:

- tex supports more formats (fine, we've already decided on which formats
  we can blit)
- FMT6_10_10_10_2_UNORM instead of FMT6_10_10_10_2_UNORM_DEST.
- FMT6_A8_UNORM for PIPE_FORMAT_A8_UNORM.

For 1555 support we get another difference between tex and color, and we
need 2/3 of the tex differences for blit srcs, so switch it over.  This
also matches what turnip does (except it doesn't have A8 as a format).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Put R8_G8_B8_420_UNORM in the format table.

turnip will use this format, even if we don't have the freedreno side
rigged up yet.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

util/format: Add an RGB planar format for YV12, like we have for NV12.

Turnip will use this for mapping the to the corresponding hardware format,
and we could also extend mesa/st to avoid adding extra samplers for
external image sampling of YV12 on freedreno.

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx+: Add support for the R8G8_R8B8 and G8R8_B8R8 formats.

This means that EGLimage imports of YUYV and UYVY can use a single sampler
instead of having to create RGBA8 and RG8 samplers mapping the same
texture.  The swizzles are the same as turnip uses, and it makes the
piglit tests for these formats happy.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Add some more 16-bit rgb/rgba swaps to our format tables.

BGR565 is chosen for GL_RGB565 and BGRA4 for GL_RGBA4, so it doesn't
change GLES's behavior.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Drop texturing support from other scaled formats.

GLES doesn't expose these texture formats, and we don't want to start
exposing them when turnip starts using our tables.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Add support for EXT_texture_sRGB_R8/RG8.

Noticed to be missing when comparing format tables between turnip and
freedreno.  Passes dEQP-GLES31 fine.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: disable vertex fetch support flag for b8g8r8a8_srgb.

We don't have an srgb decode flag on vertex fetch, r8g8b8a8_srgb isn't
supported either, and GL doesn't ask for vertex fetch for this anyway.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>

freedreno/a6xx: Add int/scaled/snorm vertex formats to match turnip.

Noticed because the vertex formats didn't match up between the two.  GL
doesn't actually look for these vertex formats, though.

Having these marked as supporting texture or color could theoretically
change things I think.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13046>