d3d12: Compute UUIDs required by external objects extension

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Reviewed-by: Giancarlo Devich <gdevich@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

d3d12: Store the rest of the device IDs in the screen

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Reviewed-by: Giancarlo Devich <gdevich@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

gallium, mesa: Support setting timeline semaphore values

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

mesa: Implement ImportSemaphoreWin32NameEXT

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

mesa: Support importing D3D12 fences as timeline semaphores

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

gallium: Add a new fence type with a pipe cap to indicate it can be imported

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

driver_noop: Remove infinite recursion from create_fence_win32

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

gallium: Add 'name' field to Win32 semaphore import

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

mesa: Implement ImportMemoryWin32NameEXT

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

gallium: Add a 'name' field to winsys_handle

Win32 memory objects can be imported by name (const void *
that will be interpreted as const wchar_t *)

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

mapi: Add more EXT_external_objects_win32 functions/enums

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

mesa: Support D3D11/D3D12 memory imports

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17446>

util: Shut down Perfetto before driver unload

Shut down Perfetto before unloading the driver to fix a crash caused by
an internal Perfetto thread that kept running after dlclose() took away
its code.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4909
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17454>

subprojects: Roll Perfetto to v27.1

Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17454>

ci: uprev vkd3d-proton to 5b73139f182d86cd58a757e4b5f0d4cfad96d319

This contains some Device Generated Command tests that will be
useful for an experimental RADV implementation.

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/17463>

zink: reject swizzled format blits

e.g., R8G8B8A8 -> B8G8R8A8 is invalid, so use u_blitter

fixes (various gl configs):
KHR-GL46.blend_equation_advanced*

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17523>

panfrost: Use early-ZS helpers

Remove the previous compile-time early-ZS implementation and replace it with the
decoupled early-ZS implementation. This uses more efficient settings in some
cases (depth/stencil tests always passes or do not write), and fixes the
settings used in another case (alpha-to-coverage enabled with an otherwise
early-ZS shader.)

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Closes: #6206
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17428>

panfrost: Add zsa->zs_always_passes flag

If we know ahead-of-time that depth/stencil testing will always pass, it's
better to use weak_early than force_early. However, if depth/stencil testing
could fail (discarding pixels), we'd rather use force_early. Determine which
case we're in at CSO create time.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17428>

panfrost: Unit test early-ZS helpers

The new early-ZS helpers are pure functions, leaf nodes of the call graph, and
implemented with a different algorithm from the "oracle" table of correct values
for various combinations of states. Further, incorrect settings often still pass
CTS while causing game bugs or inefficiencies. That combination makes the
helpers an excellent candidate for unit tests. Add some.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17428>

panfrost: Add decoupled early-ZS helpers

Bifrost (and Valhall) separate early-ZS configuration into two fields: when does
the depth/stencil buffer update happen? and when are pixels killed by the
depth/stencil tests? The driver separately configures these to occur early
(before the shader executes) or late (after the ATEST instruction executes at
the end of the shader). Early tests are generally more efficient, but various
combinations of API state and fragment shader properties can require late
updates and/or late kills for correctness. Determining how to configure these
fields is nontrivial.

Our current implementation (on Bifrost) configures these fields at fragment
shader compile time and bakes the settings into the RSD. This is both wrong
(using early testing when late testing is required) and suboptimal (using late
testing when early testing would suffice). We need to defer this configuration
until draw time, when we know rasterizer and Z/S state.

Reclassifying at draw time (as we currently do on Valhall) would be expensive,
especially with the extra terms added in here. To cope, decouple the shader
classification from the draw-time configuration. Since there are only a few bits
of draw state involved, this implementation just calculates all possible states.
Then the draw time classification is just indexing into a lookup table.

The actual algorithm used to classify is written with correctness and clarity in
mind. Unlike the current classification algorithm (which tries to match what the
DDK does, poorly), this algorithm embeds its proofs of correctness.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17428>

panfrost: Fix shader_modifies_coverage on Valhall

Alpha-to-coverage should set this flag. This is a hardware change since Bifrost.

Fixes: 26d339ef8ac ("panfrost: Generate Valhall Malloc IDVS jobs")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17428>

pan/va: Stall after ATEST

In theory this wait is required for correct behaviour of discarded threads with
ATEST. Mesa usually waits before the instruction after ATEST, so this wait will
get optimized out by va_merge_flow, but as our scheduler gets more sophisticated
this could become an issue.

Let's stay on the safe side and insert the recommended wait.

No shader-db changes.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17428>

pan/bi: Require ATEST coverage mask input in R60

In theory, ATEST can take any combination of registers for inputs.
Experimentally, however, ATEST requires the coverage mask in R60. This avoids
regressing the following dEQP tests, which write their coverage mask with
pixel-frequency-shading but without writing to the depth/stencil buffer.

dEQP-GLES31.functional.shaders.sample_variables.sample_mask.discard_half_per_pixel.*

This issue is known to affect both Mali-G52 (v7) and Mali-G57 (v9). I am unsure
if this is a silicon bug or just an obscure implementation detail.

No shader-db changes.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17428>

panfrost: Simplify sample_shading

Nos that glsl_to_nir is setting sample_shading_enable whenever FB fetch
is used, we don't need to duplicate it here.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

intel/fs,anv: Rework handling of coarse and sample shading

Now that this information is accurately gathered by spirv_to_nir, we no
longer need the hack.  We just need to fix up the way we handle some of
the key bits.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

intel/fs: Simplify persample_dispatch

Thanks to the previous commit, we no longer need this check.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

glsl/nir: Set sample_shading if a FS output ever shows up as an rvalue

If framebuffer fetch is used, we have to enable sample shading because
the fetched framebuffer value is per-sample.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

nir/gather_info: Stop gathering uses_sample_shading

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

spirv/nir: Set uses_sample_shading from spirv_to_nir

We don't really want to base this on a late nir_gather_info for two
reasons:

 1) The Vulkan spec says that if a sample-qualified input, SampleID, or
    SamplePosition are in the entry-point's interface, you get
    per-sample dispatch.  This means we really should gather this
    information before dead-code has a chance to delete anything.

 2) We want to be able to add nir_intrinsic_load_sample_pos intrinsics
    as part of lowering passes without causing per-sample interpolation.
    This means nir_gather_info needs to stop gathering it.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

glsl/nir: Set uses_sample_shading from glsl_to_nir

We don't really want to base this on a late nir_gather_info for two
reasons:

 1) The GL spec says that any static use of a sample-qualified input,
    gl_SampleID, or gl_SamplePosition causes per-sample dispatch.  This
    means we really should gather this information before dead-code has
    a chance to delete anything.

 2) We want to be able to add nir_intrinsic_load_sample_pos intrinsics
    as part of lowering passes without causing per-sample interpolation.
    This means nir_gather_info needs to stop gathering it.

For 1, this doesn't actually get us quite there as GLSL IR may have
deleted something already.  However, it does get us closer.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

intel/fs: Use nir_lower_single_sampled

This lets us drop demote_sample_qualifiers as well as a back-end check
for key->multisample_fbo.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

nir: Add a pass for lowering shaders to single-sampled

On Intel, we have to do this because we can't ask for the per-sample
barycentrics without setting the per-sample dispatch bit or the GPU will
hang.  However, nothing we're doing in this pass is Intel-specific and
it may be a useful optimization for someone else so we may as well make
it a generic NIR pass.  This version actually does a bit more than the
current brw_nir_demote_sample_qualifiers() pass as it also handles
pre-nir_lower_io interp_dref_at* as well as a couple system values which
we can easily constant-fold.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

intel/fs: Use shader_info::fs::uses_sample_shading

NIR constructs this information for us as part of nir_gather_info these
days so we can simplify our logic a bit.  This will also let us be more
correct once we move uses_sample_shading scraping earlier.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

radv: Set uses_sample_shading for copy shaders

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

intel/blorp: Set uses_sample_shading for MSAA blit shaders

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>

panfrost: Protect pandecode by a mutex

Pandecode is not thread-safe (for a large number of reasons) and does not even
try to be. This is a problem when tracing (or just using PAN_MESA_DEBUG=sync)
multithreaded applications. The most common symptom of the problem are assertion
failures deep in the red-black tree implementation, which is not thread-safe.

Just protect the whole thing by a "in pandecode?" mutex, since this is not
performance sensitive code and we don't really care about the extra
serialization incurred. As pandecode does not recurse into itself, we may simply
lock at the beginning and unlock at the end of each entrypoint in pandecode,
which is thread-safe regardless of how pandecode is used. A few entrypoints are
refactored to avoid early returns to keep the lock/unlock calls in obvious
visual pairs.

Fixes flakes when running the CL CTS with PAN_MESA_DEBUG=sync like we would in
CI (e.g: events.event_flush)

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Tested-by: Icecream95 <ixn@disroot.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17409>

panfrost: Use implementation-specific tile size

The physical tile buffer size (and hence the maximum available tilebuffer size)
are implementation-defined. Track this information on the device so we can
correctly select tile sizes, instead of hardcoding the value for Midgard.

Implementation values are pulled from the "Tile bits/pixel" row of the public
Mali data sheet [1]. That row lists the maximum number of bits available for a
pixel given the maximum tile size and pipelining. For currently supported
hardware (v9 and older), that maximum tile size is 16x16. So those values should
be multiplied by (16 * 16 * 2) / 8 to get the physical size in bytes.

This may improve Bifrost/Valhall performance on workloads using multiple render
targets. It also gets us ready for the dazzling array of tile sizes available
with v10.

[1] https://developer.arm.com/documentation/102849/latest/

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17432>

panfrost: Make pan_select_max_tile_size O(1)

Separate out "calculating the size of each pixel", "selecting a tile size", and
"calculating the colour buffer allocation". Then implement the middle (selecting
a tile size) with a simple constant time expression, rather than a loop. There's
a bit of related clean up in here.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17432>

pan/va: Handle BIFROST_MESA_DEBUG=nosb

For debugging flakes that might be caused due to wrong scoreboarding.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17223>

ir3: Fix the no-emitted-vertex condition emission in geom lowering.

The if statement we insert would insert a new block before the end block
(and remove the old pre-end-block).  If the new block ended up later in
the HT due to its pointer's hash value, you'd emit another copy of the if
statement after the last one.  I saw this happen up to 4 times in testing.
The worst case would be if all those additions and removals ended up
reallocating the HT, at which point we might use-after-free.

Fixes inconsistent shader-db results with geometry shaders.

Cc: mesa-stable.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17501>

pan/va: Print instructions with pack assert fails

va_pack asserts a large number of invariants about the instruction being packed.
If one of these fails (due to an invalid instruction), it's helpful to inspect
the failing instruction, as it may not be apparent in a large shader. Pass the
instruction through with all the assertions in va_pack for easier debugging.

Now assertion failures when packing are easier to debug:

   Invalid invariant lo.type == hi.type:
       = STORE.i32.flow2.wls br5, br2, wls_ptr[1], byte_offset:0

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17224>

pan/va: Use invalid_instruction in more places

By passing the instruction pointer through the packer, we can get better error
messages with invalid_instruction.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17224>

pan/va: Dump unencodable instructions

When we assert out due to certain invalid encoding, it's helpful to know what
instruction is causing the failure, since it may not be obvious from the
assembly for large shaders. Now we get nice errors when failing:

   Invalid opcode:
      br0 = VAR_TEX.f32.flow8.store.skip.lod_mode.center , texture_index:0, varying_index:0

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17224>

panfrost: Only key points to point coord origin

Apparently, the point coord origin within a batch can change with Gallium
(seemingly even with GLES? where that's impossible at an API level?) but that
doesn't matter if we're not drawing points. This might have to do with internal
Gallium CSOs (e.g. u_blitter). Issue noticed in SuperTuxKart, which was getting
state change flushes. Performance on one level on a Valhall GPU improved from
26fps to 29fps.

Fixes: 3641dfe4367 ("panfrost: Flip point coords in hardware")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17430>

panfrost: Revert provoking vertex assertion

b6a30b72ab3 ("panfrost: Implement provoking vertices on Valhall") added an
assertion that every draw selects a particular provoking vertex. The intent was
to ensure provoking vertex selection actually happened. Unfortunately, the
assertion is too strong, as the provoking vertex is irrelevant for some (most)
draws. For those, we don't *want* to commit to a particular provoking vertex for
those to avoid flushing.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17430>

glx: Fix drawable refcounting for naked Windows

driFetchDrawable is only ever called from the MakeCurrent path, which
means it has to handle the case of pre-GLX-1.3 Windows being named as
the drawable. When it finds the drawable in the hash, it increments its
refcount before returning it, so for a GLXWindow it would be 2 on first
return, one from glXCreateWindow and one from glXMakeCurrent. But when
it does not find the drawable and creates one for the naked Window, the
reference count on first return would only be 1. As a result, if this
context was then ever bound to a different drawable, the old Window's
DRI drawable state (like the back buffer) would be destroyed.

Fixes piglit's glx-multi-window-single-context and glx-make-current for
a variety of drivers.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6713
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17479>

intel/compiler: print shaders after nir_remove_unused_varyings

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17516>

etnaviv: tex_desc: remove descriptor patch TODO comment

There is nothing more TODO here. With softpin, which is available on all
GPUs using texture descriptors, there is no need for the kernel to patch
the descriptor, as the proper GPU virtual address is filled in by userspace.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17448>

etnaviv: tex_desc: make error handling more consistent

There already is a error handling label to free the sampler view
struct and return failure. Consistently use this label to make
error handling more uniform.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17448>

etnaviv: add texture descriptor suballocator

Texture descriptors currently waste a massive ammount of memory, as every
one is allocated via a separate BO. As the allocation granularity of the
kernel is 4KB and the descriptor is only 256B, 93.75% of the allocated
memory is wasted.

Add a simple suballocator for the texture descriptors, to allocate multiple
ones out of a single kernel BO. This isn't perfect, as freed slots in the
suballocated resource are not reused, but worst-case we end up with the
same waste as we had before. This also potentially improves efficiency at
the kernel side, as this reduces the number of BOs needed for the sampler
views in each submit.

As the BO is now used by multiple descriptors, avoid syncing with the GPU
via the cpu_prep/fini calls, as to not introduce stalls between pending
rendering and new descriptors being filled. This is safe, as each
suballocation slot is only used once, so newly filled slots are certainly
not in use by the GPU.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17448>

etnaviv: move dummy BOs to screen

The dummy texture descriptor and the dummy render target relocs are not ever
changed by a context operation, so we can save some space by moving them to
the screen and potentially share them and the BOs backing them between
multiple contexts.

Also don't hold two pointers to the same BO, one in the reloc and one raw,
but always just use the reloc one.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17448>

zink: add missing guards around `have_{ext}`

Signed-off-by: Eric Engestrom <eric@igalia.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17466>

zink: drop unused VkPhysicalDevicePortabilitySubsetPropertiesKHR

Signed-off-by: Eric Engestrom <eric@igalia.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17466>

zink: fix portability_subset usage after rename from EXTX to KHR

Signed-off-by: Eric Engestrom <eric@igalia.com>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17466>

r600/sfn: emulate pmr::monotonic_buffer_resource if needed

libc++ does not yet implement the c++17 monotonic_buffer_resource,
so emulate it by doing normal allocations that are cleaned up
when the resource is destroyed.

v2: - Use C include and version without namespace aligned_alloc
    - add include for sstream needed with clang++ (maurossi)

Closes: #6836
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17452>

r600/sfn: lower CLIPVERTEX to clip planes

With that most piglits for compatibility contexts are
passing, so enable higher compatibility profile support.

v2: fix formatting ahd fallthrough tag (Filip)

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17484>

r600/sfn: Add support for fdph

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17484>

r600/sfn: Never consider an op with register dest as dead

Another hot-fix: when a local register is written to, it is
actually unlikely that the value is never used, so just make
sure that this is never done.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17484>

r600: limit loops when trying to merge alu groups

On Cayman bank_swizzle[4] is never counted up, so add an
additional condition to make sure the loop is finished
at one point. This is a hot-fix, the logic below should be
improved.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17484>

pvr: Implicitly assert that the correct sub-command type is present

Now that we have separate C types for the different sub-command types,
we can require a pointer to that type to be passed into functions
which expect the current sub-command to be of a specific type.

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17458>

pvr: Split out unioned structs from struct pvr_sub_cmd

This is a simple optimization to make type-specific uses of struct
pvr_sub_cmd slightly less verbose.

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17458>

pvr: Implement vkCmdDraw API.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17487>

v3d,v3dv: stop copying and pasting the translate_swizzle helper

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17509>

v3d: stop using a smaller texture limit in OpenGL

The compiler has improved significantly since we found this issue
and this is no longer required.

Notice that because we are increasing the number of samplers
supported beyond what we can loop unroll (currently capped at 16),
some piglit tests that test the maximum number of samplers supported
start to fail because they use indirect indexing on a sampler array
and we don't support that (previously the indirect indexing was
removed by loop unrolling). This is a bug in tests which the
GLSL linker detects, failing to compile the shaders.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17509>

v3d,v3dv: stop hardcoding various image limits

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17509>

v3dv: clean up get_internal_type_bpp_for_image_aspects

Also, remove the FIXME to pre-compute this in images. We only use
this helper from copy/clear operations where we may be working
with a compatible framebuffer format instead of the original image.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17509>

nine: replace ulimit with sysconf call

__UL_GETOPENMAX seems to be glibc specific and not portable.
In glibc’s sysdeps/posix/ulimit.c it is assigned the return
value of sysconf(_SC_OPEN_MAX). So use the latter in the first place.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5176
CC: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17471>

iris,anv: correct the max thread number for DG2+

Correct the max thread number for DG2+ platforms according
to below bspec.

Ref: Bspec: 47202

Cc: mesa-stable
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chuansheng Liu <chuansheng.liu@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17506>

nir/opt_algebraic: Optimize [ui](add|sub)_sat with 0.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17468>

nir/lower_io: Fix array length of buffers larger than INT32_MAX.

Before, if the ssbo is too large this would always return 0.
Also, this code is easier to optimize, so the common case of offset 0
and pot stride results in one ushr instead of 5+ instructions.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17468>

ir3: Implement [iu]sub_sat.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17468>

r600: Lower uadd_sat/usub_sat.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17468>

zink: Lower uadd_sat/usub_sat.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17468>

nir_to_tgsi: Lower uadd_sat/usub_sat.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17468>

util: Don't block SIGSEGV for new threads

SIGSEGV is used by Vulkan API trace layers to track user changes in
device memory mapped to user space. Now with drivers such as Zink, GLES
applications are translated into Vulkan API calls and therefore it is
possible to be tracked by Vulkan api trace layers.
Blocking SIGSEGV hinders one of the memory tracking mechanisms used by
such layers.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17273>

v3dv: fix comment for point_sprite_mask filed in shader key

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17486>

anv: skip flush/invalidate faster

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17481>

intel: protect against empty invalidate ranges

It's legal for an application to call vkInvalidateMappedMemoryRanges()
/ vkFlushMappedMemoryRanges() with zero sized ranges.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: b91971c240d3 ("anv: use the right helper to invalidate memory")
Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6852
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17481>

radv: add a dynamic vertex format cache.

With dynamic vertex bindings the vertex format lookups are a lot
more frequent (vs being baked in the pipeline). Add a simple lookup
cache using a dynamic array to keep track of the hw values, and
avoid repeated translation.

This also reduces the memset to just the bitfields since all
the others will be overwritten.

Seen in perf traces gputest gimark with zink on radv.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15846>

anv: add a comment about handling buffer view swizzles on gfx7

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17385>

anv: validate image view lowered storage formats for storage

Ensure that if we have swizzle on the initial format, that the
component bits are identical with the lowered format.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17385>

anv: deal with isl format swizzles for buffer views

For some formats like VK_FORMAT_B5G6R5_UNORM_PACK16, we have no direct
matching HW format. We can support it by swizzling.

We already apply those swizzles for image views. We just forgot to
deal with buffer views.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6235
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17385>

docs/u_trace: Add documentation for *_GPU_TRACEPOINT envvars

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16781>

freedreno: Add FD_GPU_TRACEPOINT envvar to toggle tracepoints

All tracepoints are enabled by default.

Example:
 FD_GPU_TRACEPOINT=-flush_batch

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16781>

freedreno: Refactor tracepoints generation to reduce duplication

This way we will not need to repeat arguments for "start" and "end"
tracepoints.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16781>

freedreno: Add the rest of tracepoints with start/end to perfetto

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16781>

turnip: Add TU_GPU_TRACEPOINT envvar to toggle tracepoints

All tracepoints are enabled by default.

Example:
 TU_GPU_TRACEPOINT=-sysmem_clear

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16781>

turnip: Refactor tracepoints generation to reduce duplication

This way we will not need to repeat arguments for "start" and "end"
tracepoints.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16781>

util/u_trace: Fix iteration over config_control

Fixes: e1811af75d5b9385f9811724557a81570d0ef98e
("util/perf: add options to enable/disable tracepoints")

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16781>

ci/freedreno: Mark an occasional flake pass that happens on a530.

This MR just saw it happen for the second time I've noticed.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17470>

loader: Just include driconf.h instead of redefining its values.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17470>

egl: Just include driconf.h instead of redefining its values.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17470>

kopper: Respect the vblank_mode env var.

We were defaulting to a swap interval of 1, but we can follow dri2/dri3's
lead and respect the driconf var.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17470>

glx: Unify dri2/dri3 vblank_mode logic.

We need it from kopper, as well.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17470>

kopper: Use the swap interval that was set at swapchain creation time.

We need to track what the caller has given us for swap interval, and use
that to set the present mode at startup.

Fixes incorrect vblank syncing in apitrace's glretrace, which sets the
swap interval to 0 before the swapchain is made.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17470>

kopper: Fix the return value of kopperSetSwapInterval().

Compare to dri2SetSwapInterval() and dri3_set_swap_interval()
implementations of the same method.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17470>

etnaviv: Implement TXD

Fill in support for TXD instruction which emits shader TEXLDD opcode.

Signed-off-by: Marek Vasut <marex@denx.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17500>

etnaviv: Rename etna_emit_tex() args

Rename the args from low_bias/compare to src1/src2, since they
are used for different purposes depending on the texture load
type. No functional change.

Signed-off-by: Marek Vasut <marex@denx.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17500>

venus: GetPhysicalDeviceSparseImageFormatProperties to match sparse support

Test: dEQP-VK.api.info.sparse_image_format_properties2.*

Fixes: 1a7632e54b4 ("venus: add NO_FENCE_FEEDBACK perf option and disable sparse resource")

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17476>

microsoft/spirv_to_dxil: Fix discard to demote + return pass

While SPIR-V's OpKill is block terminating, the converted discard
intrinsic is not block terminating. This can lead to issues where
instruction could be placed after discard.

This patch adds an extra pass that drops all instructions after discard
before we convert discards.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17474>