util/disk_cache: detangle cache item creation from disk writing

This refactors the cache code so that we create the full cache
item in memory before writing it to disk. The result is slightly
cleaner code and the ability to share this code between the single
and multi file cache implementations.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9443>

util/disk_cache: separate file reads from cache item validation

This will allow us to share validation between the single file cache
and the multifile cache.

It also reduces the number of malloc() calls.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9443>

util/compress: make compression function inputs const

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9443>

lavapipe: add support for KHR_buffer_device_address.

Adds the missing magic lowering pass, and stores the pmem
pointer for easier returning to user.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9616>

lavapipe: only init immutable samplers for correct types.

This is ported from anv, and it needed to stop crashes with
the buffer_device_address CTS tests

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9616>

gallivm: convert packing to uint64 not double

Fixes some issues with later inttopotr casting.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9616>

ir3: fix oob access to regs array for getbuf,getinfo,rgetinfo

Since they have zero source registers, src->regs[1] is out of bounds.
It probably wasn't able to cause any harm, but it's always better
be safe.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4209

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9609>

ci/freedreno: run freedreno jobs on any change in src/freedreno/

Without a way to exclude specific subdirectories, keeping the list
of all subdirectories which need to be tested is error prone.
It's better to run CI slightly more often than to miss it when
it is needed.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9609>

ci/freedreno: Switch the piglit testing to the new piglit runner.

Getting piglit to fit onto our test devices was proving difficult, and we
need the ability to handle flakes, so switch to the rust piglit runner
that @pepp wrote as part of the deqp-runner repo which gives us flake
detection, sharding across boards, fractional runs, and almost half the
runtime.

It doesn't handle piglit subtests yet, but if you can't run piglit's
python on your devices because it's too bloated and unstable, this is a
way forward.

Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9468>

ci: Bump deqp-runner to v0.6.3.

This adds a piglit runner that we'll be using on freedreno, better
xfail/unexpectedpass handling for some cases (Missing->Pass, Warn->Warn),
better error logging for the Missing issue we've had in CI, a fix for a
source of Missing, and a pile more unit tests.

As a result, we now catch that the Missing results for a bunch of our
tests have gone away.

Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9468>

aco: Use ASSERTED to avoid unused variable warning.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9632>

pan/bi: Use correct enum type for NIR intrinsics.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9632>

ci/panfrost: Disable t860/radeonsi testing while the runners are struggling.

These jobs keep getting stuck, so disable for now.  Info on what is going
on is at
https://gitlab.freedesktop.org/mesa/mesa/-/issues/3437#note_841920 (the
grunt and stoney runners are maybe still operational at the moment)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9636>

zink: unset last_vertex_stage_dirty after applying it

turns out that having extra flags for stuff isn't that useful if you never reset them

Fixes: 9b8c121917d ("zink: flag shaders as needing update when clip_halfz changes")

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9633>

lavapipe: fix primitive-restart for uint8 indices

I forgot to add an implicit primitive-restart index for uint8 while
adding support for this initially. But we also have a nice little
helper to simplify this, so let's use that instead of open-coding.

Fixes: 6526e6d11b8 ("lavapipe: implement VK_EXT_index_type_uint8")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9622>

zink: add missing null check

mismerged from an earlier patch...

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9623>

CI: Trigger Windows builds for llvmpipe & Vulkan changes

As Windows now builds llvmpipe and lavapipe, we should be sure to run
their builds when these change.

Acked-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9518>

freedreno: Fix fd_fence_finish()

fd_fence_create_unflushed() is called with the driver context, but
fd_fence_finish() is called with the threaded/frontend context, which
is what threaded_context_flush() expects to be passed.

Fixes: c4e5beef076 ("freedreno: threaded_context async flush support")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9627>

intel: Simplify few version checks involving G4X

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9608>

intel: Remove GEN_IS_G4X macro

GEN_GEN and GEN_VERSIONx10 macros provide a consistent way to do platform
version checks. We can avoid platform specific macros.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9608>

intel: Simplify version checks involving haswell

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9608>

intel: Remove GEN_IS_HASWELL macro

Use GEN_VERSIONx10 == 75 check in place of GEN_IS_HASWELL macro.
GEN_GEN and GEN_VERSIONx10 macros provide a consistent way to do platform
version checks. We can avoid platform specific macros.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9608>

freedreno/a5xx: Fix the max texture buffer size.

The GLES minmax is 65536.  The blob vulkan exposes 65536 on both a5xx and
a6xx, but try just doing the same as we do for a6xx.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9617>

freedreno/a5xx: Fix the texel buffer alignment requirement.

Info comes from the a540 vulkan blob driver minTexelBufferOffsetAlignment.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9617>

ci: Run meson tests in strace if it's available and can be used

Keep the strace logs in job artifacts for tests which timed out.

This can be useful for figuring out why a timeout occurred.

strace cannot be used in jobs where ASAN is enabled, because ASAN's
leak checker also uses ptrace(), which isn't possible within strace.

Suggested-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9472>

ci: Add strace to the x86_build docker image

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9472>

aco/tests: Use _exit in child process

Since the child process doesn't call exec(), exit() attempted to run
atexit handlers registered by the parent process. This could result in
the child process hanging in exit() if there were still disk cache
threads alive when the parent process called fork(). (The CI runners
hit this multiple times when running tests in strace)

Fixes: 6a246f5c6d51 "aco/tests: Fix deadlock for too large test lists"
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9472>

ci: Move /usr/bin/time check from meson test wrapper to build script

In jobs where the wrapper isn't used, this leaves the
<build directory>/meson-logs/testlog.txt filename unchanged.

Also prepares for using different wrapper scripts in different jobs.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9472>

ci: Drop SIGINT handling from meson test wrapper script

Not needed, I had misread meson code.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9472>

ci: Move meson-build.sh to meson/build.sh

A little less clutter in the top .gitlab-ci directory.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9472>

d3d12: Really handle null constant buffers

Previous attempt was checking the wrong pointer...

Fixes: eb27db86 ("d3d12: Handle null constant buffers")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4362
Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9349>

d3d12: Handle is_new_style_shadow comparison filtering

Note: the tex instruction is modified in place, so the correct
number of expected result components needs to be cached before
the shadow state is modified

Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9349>

d3d12: Move descriptor pools to screen, and add lock

Surfaces can be shared across contexts, and can even outlive the
original context that created them, so move the pools to the screen.
Since they no longer belong to a single context, they need a lock now.

v2: Samplers moved back to the context
v3: Fix Linux build

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3812
Reviewed-by: Bill Kristiansen <billkris@microsoft.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9349>

vulkan: add missing vk_shader_module.c/h includes to Makefile

Fixes: 76078ed9fe3 ("vk/util: add unified shader module struct/functions")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9620>

zink: cache bufferviews

we can cache bufferviews just like we do surfaces, so let's do that

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: properly handle null bufferview descriptor states

now that we have this state properly stored we can use it for more accurate
hashing and caching

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: improve debug asserts for samplers/images during descriptor updates

these are now more accurate for null bufferviews/imageviews

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: update null descriptor hashes to reflect current descriptor states

also add a bufferview hash since we use these now

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: add helper function for getting a resource for a descriptor

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: massively beef up batch tracking for shader images

the struct for these isn't allocated, so we need to ensure that the objects
in it are explicitly tracked on batches when they're used

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: add extra batch tracking for sampler views

we're going to start invalidating and rebinding resources, which means
we need to be extra sure that we have our lifetimes in order here

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: simplify bufferview and imageview descriptor state hashing

now that we have the hash for the internal objects available at all times, we
can directly reuse this instead of computing it each time, giving us more
accurate hashes as well as saving us cpu cycles

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: store bufferview hash to bufferview struct

this lets us do a lot of things more precisely for bufferviews

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: explicitly use zink_surface objects for sampler/image view objects

we can reuse this codepath instead of NIHing it

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: use an explicit zink_buffer_view struct

this gives us an object to work with similar to zink_surface

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: add util function for checking whether shader descriptor is buffer from program

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: add util function for checking whether a shader descriptor is a buffer

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: add some asserts for pipeline barriers to check renderpass state

it's illegal to emit barriers during a renderpass

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

zink: handle cached descriptor set punting

if an active set gets invalidated, it needs to be punted from the active
hash table so it doesn't get reused

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9543>

CI: Try really hard to get updated Windows TLS certs

Windows doesn't actually distribute a full TLS CA certificate store, but
pulls them in over time with Windows Update. Try to prime it by manually
pulling the certificates and installing them.

This bumps the Windows tag to force a rebuild.

Acked-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9618>

zink: track resource_object usage instead of resource usage

the resource object is now a state tracker for the inner resource_object,
so it can safely be destroyed even if the backing resource is still in use

this also requires updating various things to keep descriptor states/caches working

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9544>

zink: split out backing resource object create/destroy

moving towards these being able to exist semi-independently...

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9544>

zink: move resource internals to a separate struct

this starts the path towards being able to replace the backing objects for
a resource

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9544>

zink: break out batch id finding for resource usage into util function

this is generally helpful for figuring out the last batch a resource was
used for

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9544>

zink: avoid looping for non-ubo descriptor updates based on set usage

if we know that the descriptor set is cached and already in use by a given
batch, then we also know that all the resources in the set are tracked, which
means that we can skip over some looping  during descriptor updates which would
otherwise be used to add tracking for sampler/image views/states

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9567>

zink: add batch usage flags for sampler views/states and desc sets

this massively cuts down cpu time for hashing and lookups, resulting in
a noticeable performance increase

the sampler_state batch usage for now is just being set for future use when
it will be leveraged to permit immediate deletion

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9567>

radv: Use correct DCC compressed block size for sampling.

Don't need to change the pre-GFX9 samplers because this option doesn't
vary there.

Fixes: f848f2adfae ("radv: Use ac_surface DCC settings for shareable images.")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4455
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4425
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9611>

genxml: Make 1-bit L3$ config register fields bool on Gen7

Otherwise, they look like booleans but, if you put a value other than
0/1 in them, the GenXML generator code will explode.

Fixes: b6875b0094c "anv: Drop has_slm in emit_l3_config for gen11+"
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9614>

turnip: implement intrinsic_vulkan_resource_reindex

Descriptor arrays are continuous, so it's just an addition of offset.

Fixes test:
 dEQP-VK.spirv_assembly.instruction.compute.variable_pointers.dynamic_offset.select_descriptor_array

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9495>

ci/freedreno: Add three more a5xx flakes from the last day.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9575>

android: i965: Rename files with "intel_" prefix to "brw_"

The necessary changes for Android build were missing in Makefile.sources

Fixes: 9d95e1bd79a8 (i965: Rename files with "intel_" prefix to "brw_")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9576>

anv: use common interfaces for shader modules

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9508>

tu: use common interfaces for shader modules

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9508>

v3dv: use common interfaces for shader modules

squashed changes from Alejandro Piñeiro <apinheiro@igalia.com>:

Add call to vk_object_base_init on internal shader_module: we have
some cases where internally we have some shader modules that we don't
create through CreateShaderModule, so in this case we need to manually
call base_init. Not sure why this wasn't needed before.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9508>

radv: use common interfaces for shader modules

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9508>

lavapipe: use common interfaces for shader modules

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9508>

vk/util: add a util macro for initializing stack vk_shader_module structs

radv does a lot of this, so having a central dispatch point will be useful in
case changes are made to this struct in the future

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9508>

vk/util: add unified shader module struct/functions

there's some extra logging stuff dumped into here to match functionality,
eventually that should also be consolidated into vk_util.c

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9508>

lavapipe: add host query reset

This just resets the queries on the host size.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9594>

lavapipe: fix writing availability for queries.

If the availability has to be written it needs to be written
 to the correct place.

The host query reset tests fall over this.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9594>

lavapipe: add missing break

No effect as-is but annoying later.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9594>

anv: Set correct binding table entry count

We can use surface_count as it is to set binding table entry count since
it's already in units.

On Felix's Tigerlake with the GPU at fixed frequency, this patch
improves performance of several games:

   - Shadow of the Tomb Raider: +1.5%
   - Dota2vk: +1%
   - Dark Souls: +1%

v2: (Ken)
- Remove get_binding_table_entry_count() and use surface_count directly.

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9548>

anv: Drop has_slm in emit_l3_config for gen11+

For some gen12+ platforms, L3 config (cfg) can be NULL leading to a
seg-fault in emit_l3_config. But, we don't use has_slm for gen11+, so
we can just avoid declaring the variable.

Reworks:
 * Drop has_slm variable for all gens (suggested-by Jason)

Ref: 633dec7163e ("anv: Set L3 full way allocation at context init if L3 cfg is NULL")
Ref: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9534
Fixes: 581e68bc99b ("anv: move L3 config emission to genX_state.c")
Signed-off-by: Jordan Justen <jordan.l.justen@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/9589>

anv: Remove redundant #if checks

Patch removes redundant #if checks and unused code inside another #if
block.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9604>

anv: Lower ViewIndex to zero when multiview is disabled

Vulkan spec says

    If multiview is enabled in the render pass, this value will be one
    of the bits set in the view mask of the subpass the pipeline is
    compiled against. If multiview is not enabled in the render pass,
    this value will be zero.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4446
Fixes: 0db70703300 ("anv/pipeline: Add shader lowering for multiview")
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/9574>

zink: stop leaking programs

gfx and compute programs both end up with an extra ref from creation,
so they both need to lose a ref during creation after they get their
shader refs

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9470>

zink: unset ctx->program pointers when an unref destroys the object

if we destroy a program object which is currently the "active" program then
we need to unset the pointer to avoid invalid access

also unset injected tcs pointers where appropriate

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9470>

zink: return true from program ref functions upon free

this lets us unset the context pointer accurately

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9470>

zink: use internal api for first-frame fence

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9290>

zink: force explicit fence only on first frame flush

we have implicit sync hooked up for drivers now so we don't need to worry
about drawing over our frontbuffer unexpectedly

still a weird issue remaining where we miss the first frame without an explicit
fence, but I expect that will get resolved eventually by wsi

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9290>

zink: force mesa flush implicit fencing on ANV

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9290>

zink: determine whether the vulkan driver requires mesa flush wsi

this is used for ANV's implicit fencing (until we get wsi)

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9290>

zink: use a safer iteration for fb surfaces during rp init

doing a pointer iteration looks cool, but if the array is full, then it
goes out of bounds and we crash

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9542>

zink: implement a surface cache

this is a global cache for all surface objects, enabling some memory
optimizations as well as improved reuse of cached descriptors

loosely based on patches from Antonio Caggiano

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9541>

zink: ralloc screen objects

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9541>

swr: fix array-bounds warning

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9502>

swr: fix deprecated llvm 11 declaration warning

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9502>

swr: fix unused SplitString warning

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9502>

swr: Fix SWR_CONTEXT pre-declaration

Silents a warning on msvc

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9502>

swr: fix win32 intrinsics

The doc doesnt mention Index is altered when mask is null.
This is consistent with both llvm & migw implementations.

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9502>

swr: extern declaration for win32 intrinsics

This fixes compilation on msvc because llvm redefines these too
Closes #4417

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9502>

radv: rework radv_use_dcc_for_image() a bit

To make it clear that only GFX8-9 have missing DCC features.

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/9526>

radv: remove useless DCC disable check for 3D images on GFX10+

addrlib uses the S swizzle mode which disables DCC completely.

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/9526>

radv: add missing SQTT events for copy_commands2/create_renderpass2

A bunch of entrypoints were missing.

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/9560>

anv: fix memory allocation error handling

Reported by Coverity.

Fixes: 0a7224f3ff7 ("anv: group as many command buffers into a single execbuf")
Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9596>

intel: install intel_device_info

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Suggested-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9595>

frontends/va: fix protected slice data buffer read size

Read vlVaBuffer->size bytes instead of an hardcoded value.

Fixes: deb7dc82f62 ("frontends/va: handle protected slice data buffer")
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Reviewed-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9553>

radeonsi/sqtt: fix user event max size

Larger strings can corrupt rgp files.

Fixes: 5dc823304b1 ("radeonsi/sqtt: forward string markers to sqtt")
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Reviewed-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9553>

broadcom/compiler: be more flexible scheduling TMU writes

V3D 4.x allows more flexibility, so take advantage of that. Generally,
we can reorder any writes in the same sequence, so long as they are
not the sequence terminator (which must always be last, since it is
the one triggering the operation), and TMUD writes, since these must
be ordered with respect to each other.

total instructions in shared programs: 13735183 -> 13731927 (-0.02%)
instructions in affected programs: 903057 -> 899801 (-0.36%)
helped: 2358
HURT: 746
Instructions are helped.

total max-temps in shared programs: 2322020 -> 2322009 (<.01%)
max-temps in affected programs: 619 -> 608 (-1.78%)
helped: 19
HURT: 11
Inconclusive result (value mean confidence interval includes 0).

total sfu-stalls in shared programs: 31494 -> 31489 (-0.02%)
sfu-stalls in affected programs: 182 -> 177 (-2.75%)
helped: 40
HURT: 40
Inconclusive result (value mean confidence interval includes 0).

total inst-and-stalls in shared programs: 13766677 -> 13763416 (-0.02%)
inst-and-stalls in affected programs: 901343 -> 898082 (-0.36%)
helped: 2349
HURT: 746
Inst-and-stalls are helped.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9555>

broadcom/compiler: flag wrtmuc with a read dependency on last_tmu_config

Instead of using a write depdency. We use last_tmu_config to ensure ordering
of instructions participating in different TMU sequences. To this end,
all sequence terminators flag a write dependency on last_tmu_config, but
wrtmuc is not a sequence terminator, so we can be more flexible by flagging
it as a read depedency. This would prevent it to be moved into a previous
sequence (since it cannot be moved past the previous sequence terminator due
to the read depedency), but it allows it to be reordered with instructions in
the same sequence, which allows us to pair it up more effectively. Particularly,
it allows to pair up a wrtmuc with the sequence terminator of the same sequence,
turning code like this:

nop                  ; mov  tmut, r0     ; thrsw; wrtmuc (tex[0].p0 | 0x3)
nop                  ; nop               ; wrtmuc (tex[0].p1 | 0x0)
nop                  ; mov  tmus, r1

Into this:

nop                  ; mov  tmut, r0     ; thrsw; wrtmuc (tex[0].p0 | 0x3)
nop                  ; mov  tmus, r1     ; wrtmuc (tex[0].p1 | 0x0)

total instructions in shared programs: 13755738 -> 13735183 (-0.15%)
instructions in affected programs: 2510921 -> 2490366 (-0.82%)
helped: 10963
HURT: 485
Instructions are helped.

total max-temps in shared programs: 2322828 -> 2322020 (-0.03%)
max-temps in affected programs: 11303 -> 10495 (-7.15%)
helped: 608
HURT: 19
Max-temps are helped.

total sfu-stalls in shared programs: 31545 -> 31494 (-0.16%)
sfu-stalls in affected programs: 235 -> 184 (-21.70%)
helped: 62
HURT: 11
Sfu-stalls are helped.

total inst-and-stalls in shared programs: 13787283 -> 13766677 (-0.15%)
inst-and-stalls in affected programs: 2525187 -> 2504581 (-0.82%)
helped: 10989
HURT: 477
Inst-and-stalls are helped.

v2: add a comment explaining the read depdency (Piñeiro).

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9555>