nvk/drm: Restructure nvk_queue_submit_drm_nouveau()

Now that we don't need the lock, we can return directly.  Also, now that
we don't have the old UAPI, we can clean things up and make the whole
function make a bit more sense.  Also, drop some pointless braces while
we're just moving code around.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

nvk: Get rid of the tiled memory allocation helpers

These existed entirely to support shadow memory for VkImage cases where
we needed tiling.  Now that we have VM_BIND, these are no longer used so
we can drop the wrappers and just implement VkAllocate/FreeMemory again.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

nvk: Drop the device-level mutex

This existed to let us lock the memory_objects list and for handling
BO-based vk_sync waits.  We don't have either of these things anymore so
there's no need for a device-level lock.  We already have fine-grained
locks around the data structures that need them.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

nvk: Drop memory object tracking

We no longer have to pass a list of BOs to the kernel so there's no need
to track allocated memory objects like this.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

nvk: Drop command buffer BO tracking

We no longer need to provide this information to the kernel so there's
no reason to track it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

nvk/drm: Drop the push_add_bo() helper

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

nvk/drm: Drop the push_add_push_bo() helper

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

nvk: Rip out old UAPI support

This was mostly generated by

    unifdef -DNVK_NEW_UAPI=1 -m src/nouveau/**/*.h

with a few manual fixups for device features and to remove includes and
strip out the meson bits.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357>

anv: fix a leak of fp64_nir shader

Fixes: 8c4c4c3ee1a2 ("anv: Add softtp64 workaround")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9846
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25372>

util: Fix bookkeeping of linear node sizes

When creating a new node, we were clobbering the original size
requested, and use that as offset, so the node would always be full.

Fixes: 591db9a9a54 ("util: Remove per-buffer header in linear alloc for release mode")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25382>

draw/mesh: reset some user state values on mesh draws.

Otherwise previous non-mesh draws and can leave things dirty.

Fixes crashes in:
dEQP-VK.mesh_shader.ext.query.all_queries.triangles.reset_before.copy.no_wait.indirect_draw.32bit.no_availability.no_blocks.task_mesh.inside_rp.single_view.with_secondary
after other shaders have run.

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25385>

anv: don't use cmd_buffer after destroyed.

Fixes: 13b3d7f74 ("anv: Handle companion RCS in end/destory/reset code path")

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25366>

svga: use srv raw buffer for accessing readonly shader buffer

When a surface is to be created for a buffer that is to be persistently mapped,
set the UAV bind flag as well, otherwise, an extra copy of the surface will have
to be created when the buffer is later bound as shader storage buffer, and
it will then incur extra buffer copies to keep the shadow copies of the surface
in sync. But when a buffer is simultanously bound to uniform buffer and shader
storage buffer, having both srv and uav on the same resource can also be costly.
This patch also provides optimization for readonly shader buffer. In this case,
instead of creating uav for accessing a readonly shader buffer, srv
raw buffer is used instead.

Tested with yuzu tracefile included in VMware bug 3029385.

Reviewed-by: Martin Krastev <krastevm@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25381>

svga: fix immediates used in rawbuf for constant buffer

Since constant buffers can be accessed as raw buffers, indices to
the constant buffers will have to be declared as immediates.
But it is a waste to define immediates for every possible indices
to the constant buffer, we will only include immediates that are used in
the shader. But since immediate block is declared in a very early stage
of a shader, this patch will append any new immediates to the immediate
list and reemit the immediate block if needed.

Fixes assertion running yuzu

Reviewed-by: Maaz Mombasawalam <mombasawalam@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25381>

svga: restrict use of rawbuf for constant buffer access to GL43 device

Restrict use of rawbuf for constant buffer access to GL43 capable
device only.

Fixes glretrace regressions running with SW Renderer.

Reviewed-by: Martin Krastev <krastevm@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25381>

svga: fix persistent mapped surface update to constant buffer

Currently when a buffer is mapped with the persistent bit set and is
later bound as a constant buffer, the updated buffer content is not
properly updated to the constant buffer surface as the constant buffer
surface is different from the original buffer surface. Doing
a buffer copy to sync the content of the constant buffer will fix
the problem, but the buffer copy can be costly.
To properly fix the issue, instead of creating a secondary surface
for the constant buffer, the original buffer surface will be accessed as
a raw buffer.

This fixes the rendering issue running yuzu deko_basic.nro

Reviewed-by: Martin Krastev <krastevm@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25381>

ci: Upref virglrenderer

Pull in the latest changes regarding texture wrapping
modes handling.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25343>

virgl: report MIRROR_CLAMP features better

The new host version checks the support of these features better,
so report here accordingly. This fixes a number of texwrap piglit
tests on Intel.

v2: Stick to old test for PIPE_CAP_TEXTURE_MIRROR_CLAMP because
    host has to be backward compatible.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25343>

nvk: Advertise discreteQueuePriorities = 2

This is a lie since we don't do anything with queue priorities at
present but Vulkan requires us to advertise at least 2.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25383>

nvk: Advertise maxMemoryAllocationCount = 4096

This is the minimum and also what the proprietary driver advertises.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25383>

nvk: Re-order physical device limits

They now match the order in vulkan_core.h which makes it easier to find
missing entries of which there was one.  This also fixes a bug where we
were setting .bufferImageGranularity twice and we were overwriting the
correct value with an incorrect one.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25383>

tu: Remove pipeline from state

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Don't use pipeline for active stages

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Don't use pipeline for per_view_viewport

We already have the state in the command buffer.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Don't use pipeline for bandwidth validity

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Don't use pipeline for dynamic draw states

For dynamic states that are precompiled from static state, just set the
corresponding dynamic draw state directly, and keep a record of which
ones are precompiled when we go to emit states at draw time so we don't
accidentally re-emit them.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Make compute dispatch use the shader

With this, the last use of the compute pipeline is gone. We still use
the pipeline for the load state, though.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Don't reference pipeline for some draw states

These draw states are things that depend on pipeline-only state:

- The load state depends on knowing the pipeline layout, which we won't
  know for a shader that's loaded from a binary. This is going away on
  a7xx anyway, and we should be able to use the a7xx strategy of
  prefetching the descriptors in the preamble on a6xx too.
- The prim order state depends on feedback loops and raster order
  attachment access, which isn't supported at the moment.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Rewrite remaining pipeline LRZ handling

Now that the FS-specific parts are split out, the only remaining part is
the blend state part. Use the same state that we use for dynamic
blending for static blending, eliminating the last use of the pipeline
in the LRZ code. While we're at it fix a bug where dynamic blending
didn't always disable LRZ writes (even though it only mattered with a
non-conformant debug flag because we invalidated LRZ anyway).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Use pipeline feedback loop flag indirectly

Remove another use of the pipeline.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Decouple program state from the pipeline

There are a couple pieces of state that we can't calculate until we know
all of the shaders:

- The actual variants to use (i.e. whether to use safe-const variants)
- Program config and VPC draw states
- Const layout, which depends on the variants
- Whether per-view viewports should be enabled

Now that these are all combined in tu_pipeline::program, move these into
a separate struct that can be referenced directly without a pipeline.
The next step is to refactor the code filling it out so that it can be
called at draw time when given just the shaders.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Rework passing shared consts

The way this works now is awkward to map to shader objects. We don't
have the pipeline layout when "linking" shaders at draw time, so we have
to piece it together from the shaders. Store the information we need in
the shaders and piece it together.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Rewrite tessellation modes handling

Before this, we combined the modes after compiling the shaders when
constructing the pipeline. But that's a bit awkward with shader objects,
where there is no good place to put state derived from TCS and TES but
not the other stages. However, shader objects leaves us with an out:
when compiling separately, the modes must be on one of the shaders. So
instead we just copy the modes earlier, in the NIR shaders, and then get
them from the appropriate shader later. That way there is no extra
overhead when fast-linking, as there currently is, and we don't need to
create an awkward separate object just for this.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Use shader directly for VS/TCS output size and patch size

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Move FS-specific pipeline information to the shader

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Start tracking shaders independently of pipeline

We will gradually transition over users of cmd->state.pipeline and
TU_CMD_DIRTY_PIPELINE to shaders and derived state from shaders. This
just puts in place the framework to start doing that.

When importing a library with all of the shader state, we now have to
import the shaders in addition to the program, so that they are
available when we bind the pipeline.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

tu: Create singleton "empty" shaders

The most important of these is the empty FS, which should cut down on
time spent in the compiler when we fast-link a pipeline where there is
only a library with the VS and no library with the empty FS. Source
engine is known to do this.

This is also necessary for shader objects where the "empty" shaders are
never created up-front. We will use these when a NULL shader is bound.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25276>

util: Add a few basic tests for linear_alloc

Not comprehensive, but those were the ones used to work on the
previous linear_alloc changes.  Also having a test already
set up lower the barrier to add more tests for future in case of
bugs.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Remove per-buffer header in linear alloc for release mode

There's only need to keep the offset and size of the latest buffer,
so rename linear_header into linear_ctx and change the code to
keep records there.

For debug mode we still keep a header, now called linear_node_canary,
to have a magic check.  Since due to alignment we have a free space,
also keep the individual occupation of each node (offset), for
debugging.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Remove size information from child allocations

With linear_realloc() gone, there's no code that reads the size
in linear_size_chunk struct, so it can be removed.  This removes
the 8-byte overhead per child allocation and simplifies the
allocation code.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Remove linear_realloc()

Now that linear_realloc() is unused, remove it.  It is not an actual
realloc, will always allocate new memory and copy data around -- and
had a big warning about it in the documentation.

In the couple of uses we had before, the client code knew the size,
so it could be changed to perform the allocation and the copy by
themselves.  The client code keeping the size is the recommended
way here.

This will allow us remove linear_size_chunk later.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Remove usages of linear_realloc()

Note that for linear allocator, the realloc will always
allocate new memory.  In both cases that realloc was used,
the existing size was known, so we can just allocate
and do the copy ourselves.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Use an opaque type for linear context

In the linear allocation only the parent (context) can be used
to allocate new children, so let's use an opaque type to identify
the linear context.  This is similar to what's done in GC allocator.

Update the documentation and a couple of function names to
refer to linear context instead of linear parent.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Make DECLARE_LINEAR_ALLOC_* macros assume no destructors

Linear allocator doesn't support calling custom destructors to
its child allocations nor freeing individual child allocations.
So the destructor callback and the delete operator don't apply
to objects using linear allocator.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Remove size from linear_parent creation

None of the callsites took advantage of this, so remove
the feature.  This will help to a next change that will
add an opaque type to represent a linear parent.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Use linear parent to (r)allocated extra nodes

Instead of making all the nodes children of the same original
ralloc_ctx, use that context just for the first node (which
is the linear parent).  The remaining nodes are (r)allocated
using the first one.

With that change, to free/steal, it is sufficient to apply the
operation to the first node, which uses the underlying linked
list between the nodes at ralloc level.  And we can remove
the 'next' pointer in linear_header, now unused.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

util: Remove ralloc_parent from linear_header

We can reach the ralloc parent from the linear parent, since
it is what was used to allocate the linear parent in the first
place.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25280>

radv/ci: skip dEQP-VK.robustness.* on Vangogh due to weird GPU hangs

For very weird reasons, this subset hangs on VanGogh only. It took a
while to figure out that and I can't reproduce on my Steam Deck. Skip
the whole subset for now until the issue is figured out to avoid
blocking the CTS update.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24947>

ci: update to vulkan-cts-1.3.6.3

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24947>

freedreno: Include speedbin fallback in 730 chipid to fix probing

The kernel exposes CHIP_ID, which consists of:

[63:48] const 0 / reserved
[47:32] speedbin (default 0xffff)
[31:0 ] actual gpu chip ID

Fix the 730 chip id to make it probe correctly.
Keep the existing entry for the downstream KGSL driver.

This is essentially the same as Commit 6067aba9e6db ("freedreno: Include
speedbin fallback in 740 chipid to fix probing"), except for the older
core.

Fixes: cd3719be7d09 ("freedreno: Fully define a730 and a740 device properties")
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25375>

aco: reset prefetch in the correct block after removing the exit

fossil-db (navi31):
Totals from 279 (0.35% of 79332) affected shaders:
(no stat changes)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: c778803d673f ("aco/assembler: change prefetch mode on GFX10.3+ during loops if beneficial")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25312>

intel/ds: avoid dropping traces when running out of shared memory

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

anv/utrace: trace CPU on timestamp buffer readiness

Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

intel/ds: track number of tracepoint timestamp copies

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

util/u_trace: count number of tracepoints

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

intel/ds: lock submissions to u_trace_context

This allows for submissions to be outside the global anv_device mutex.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

intel: don't assume Linux minor dev node

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

anv: enable utrace timestamp buffer copies on compute engine

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

anv: move utrace flush out of backends

The next patch is going to introduce some locking that needs to happen
before the submission to the backend.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

anv: move simple shaders code to its own object

We would like to call those function from anv_utrace.c

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

anv: add simple shader support without a command buffer

Limited to compute for now. Annoyingly Gfx9 requires a binding table
block.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

anv: add a memcpy compute internal kernel

We'll use this memcpy utrace timestamp data.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

anv: fix utrace timestamp buffer copies

Fixes: 521c216efc ("anv: use COMPUTE_WALKER post sync field to track compute work")
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24744>

frontends/va: Set default rate control values once when creating encoder

Move setting the default values from getEncParamPresetH264/5 as this
function is called on each frame which would result in overwriting
values set by application.

This fixes setting HRD parameters and max_qp/min_qp when
PIPE_VIDEO_CAP_ENC_QUALITY_LEVEL is not supported.

Cc: mesa-stable
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25355>

frontends/va: Process VAEncSequenceParameterBufferType first in vaRenderPicture

The encoder is created in handleVAEncSequenceParameterBufferType and it
also sets some default parameters, so we need to make sure to handle
this buffer first because application may have already set those
parameters from earlier buffers.

This fixes setting HRD parameters with gstreamer vah264enc/vah265enc
when PIPE_VIDEO_CAP_ENC_QUALITY_LEVEL is supported.

Cc: mesa-stable
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25355>

docs: add calendar for 23.3

Normally the branchpoint would have been on Oct 11, but this is the week
before XDC and everyone will be busy so let's try not to add the
branchpoint pressure on top. Let's instead branch off on the week
_after_ XDC, on Oct 25.

Besides, 23.2 has been delayed by multiple factors, so it wouldn't make
sense to rush 23.3.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25122>

radv/ci: add more tests to the navi21 vkcts flake list

Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25325>

radv/ci: increase the parallelism of the vkcts-navi21 job

The vkcts-navi21 job currently take ~13 minutes each, but we are about
to start running the VK_EXT_shader_object tests which should
significantly increase our runtime and thus would exceed the 15 minutes.

Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25325>

v3d/shim: include new ioctl parameters

Some new parameters added recently were missed in the DRM shim.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9880
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25370>

radv: enable cache counters for RGP on GFX11

This was the last missing RGP feature on GFX11 compared to older chips.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25271>

radv: add SPM support for GFX11

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25271>

ac/spm: add support for GFX11

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25271>

ac/spm: add new segment types for GFX11

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25271>

ac/gpu_info: define AMD_MAX_WGP

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25271>

llvmpipe/linear: move to nir inputs

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24436>

llvmpipe/linear: convert to using nir for output.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24436>

llvmpipe/fs: start converting dervied state to nir based.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24436>

llvmpipe/fs: start converting interp/input paths to nir.

This input_usage_mask is going to take a bit more thinking

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24436>

llvmpipe/nir: call gather info to update inputs read properly

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24436>

gallivm: drop tgsi aos paths.

These were only used by the linear code, and that doesn't use
tgsi anymore

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24436>

llvmpipe/linear: drop tgsi path.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24436>

ci/freedreno: disable broke cheza (Adreno 630) runners

Revert ASAP as `google-freedreno-cheza` gets running again.

Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25365>

r600/sfn: Fix use of scheduled_shader vs shader

Thanks to Vitaliy Kuzmin for pointing this out.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25364>

r600/sfn: sort the uniforms of the right shader

Thanks to Vitaly Kuzmin for pointing this out.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25364>

r600: Use nir_builder helpers for load/store_shared_r600

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24762>

r600: Use more auto-generated nir_builder helpers

Use nir_umad24 instead of rolling our own.  Also, all system values
already have NIR builder helpers and we don't need to be manually
building intrinsics or rolling our own helpers.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24762>

pvr: Fix dynamic desc offset storage

The index at which the dynamic descriptor offsets were being
stored was incorrect, leading to some offsets not being stored and
thus `0` being applied as the offset to the descriptors instead.

dEQP test fixed:
  dEQP-VK.binding_model.shader_access.{primary,secondary}_cmd_buf
    .uniform_buffer_dynamic
    .{vertex,fragment,compute,vertex_fragment}
    .multiple_discontiguous_descriptor_sets
    .*_descriptor.offset_view_{,non}zero_dynamic_nonzero

Fixes: aa791961a82e ("pvr: Add support for dynamic buffers descriptors")
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25320>

pvr: Fix `for` loop itarator usage

The `i` iteration variable was being used instead of `k` when
appending new mappings, and getting the current source rect.

dEQP test fixed:
  dEQP-VK.pipeline.monolithic.image.suballocation.sampling_type
    .combined.view_type.3d.format.*.count_1.
    .{3x3x3,5x5x5,...(odd dimensions)}

Fixes: 060c3db4efd6 ("pvr: Complete pvr_generate_custom_mapping()")

Reported-by: James Glanville <james.glanville@imgtec.com>
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25321>

nvk: Set GUARDBAND_Z_SCALE_1 when Z-clipping

This fixes most of the remaining dEQP-VK.*.inverted_depth_range.* tests.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25359>

nvk: Enable dynamic clip/clamp enable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25356>

nvk: Fix depth clipping parameters

This is a pretty complete overhaul which I think may actually be correct
and there are now comments explaining why various parameters were
chosen.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25356>

nvk: Clean up viewport math

This should make everything a bit easier to read

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25356>

iris: do not mention specifically clover for OpenCL support

It's confusing, when rusticl is used.

Acked-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25287>

ci/venus: add fragment.32B_in_memory_with_vec4_s32 flake

Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25349>

anv: fix frame count reporting in INTEL_MEASURE

Report frame count at CB submit time, instead of CB build time.

Reviewed-by: Mark Janes <markjanes@swizzler.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25235>

nvk: Disable statistics around meta ops

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25353>

nvk: Lower interp_at_sample to interp_at_offset

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25353>

pvr: remove pvr_pbe_get_src_pos()

This should have been removed as part of f59c6eef168 ("pvr: Move PBE START_POS
into csb enum helpers header").

Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25328>

pvr: treat VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT as not supported

The required compiler support is missing for mutable format support, so return a
not supported error from vkGetPhysicalDeviceImageFormatProperties2() when the
VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT flag is set in the image format info
structure.

This results in the dEQP-VK.image.mutable.* conformance tests being treated as
Not Supported.

Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25241>

lavapipe: set default min sample shading to 1

this should only affect shobj handling and even then probably no effect

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25354>