amd/llvm: fix C++ compile failures

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7807>

gallium/util: allow including a few files in C++

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7807>

compiler: fix glsl_types.h compile failures when including as C++ in drivers

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7807>

ac/llvm: handle no_(un)signed_wrap NIR flags

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

ac: unify shader arguments that are duplicated

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

radeonsi: move si_llvm_compiler_shader and deps into si_shader_llvm.c

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

radeonsi: move si_build_main_function into si_shader_llvm.c

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

radeonsi: move si_create_function into si_shader_llvm.c

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

radeonsi: split ac_shader_args initialization from LLVM code

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

ac: add shader return values into ac_shader_args

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

ac: correct ac_shader_args types, remove sgpr_count

sgpr_count is unused. The size of the others is too small.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7939>

radeonsi: fix a crash in si_fence_server_sync

Fixes: 1f31a216640 - radeonsi: remove SDMA support

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8010>

aco: rename s_subb_u32 operands to borrow

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8007>

aco: fix various s_subb_u32 operands to SCC

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8007>

v3d: use job's nr_cbufs field

Instead of computing the number of color buffers in the start binning
function, just use the job's nr_cbufs, which already contains that
value.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8006>

panfrost: Fix provoking vertex selection for lines

For line primitives, the provoking vertex selection is done through the
DRAW.flat_shading_vertex field and PRIMITIVE.first_provoking_vertex must
be set to true.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7923>

panfrost: Preload SampleID when reloading multisample FBs

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7923>

panfrost: Take the number of samples into account in blend shaders

Midgard has to split the writeout instruction if the number of bits per
pixel exceeds 128. We thus need to take the number of samples into
account when creating blend shaders.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7984>

pan/mdg: Add support for multi sample iteration writeout

Some MSAA+fmt combination require writeout to be split. Right now, it
only impacts blend shaders since we only support MSAA 4x, and the only
formats that could exceed the 128bit/pixel limit in MSAA 4x are
not supported by the fixed-function blend unit. We thus rely on the
blend shader to split things properly. Things will change once we add
MSAA 8x/16x to the mix, since even the blendable formats will exceed
the 128b/pixel limit in that case.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7984>

panfrost: Fix fencing

Commit 64d6f56ad26f ("panfrost: Allocate syncobjs in panfrost_flush")
aimed at optimizing the fencing logic but it looks it also broke the
fence-based synchronization in subtle ways.

Indeed, now that the fence only waits on a single syncobj, we're not
guaranteed that all jobs queued in panfrost_flush_all_batches() will
be done when the fence is signaled, because jobs at the top level
(those stored in the batches hashmap) have not inter-dependencies.

Commit 9e397956b092 ("panfrost: signal syncobj if nothing is going to
be flushed") made this even more apparent by signaling the fence right
away if nothing was left to be drawn in the current context, thus
ignoring any of the batches left to flushed in the ->batches map.

If we want to keep relying the existing kernel APIs there's clearly no
ideal solution here. We can either go back to the original fencing
mechanism where each fence contained an array of syncobjs to be tested
or serialize jobs that have no explicit dependencies so we know the last
submitted job will also be the last one to return. The orginal approach
has proven to add quite a significant overhead (caused by the amount of
ioctls and the time spent in kernel space to gather dma fences attached
to those syncobjs and test them). So let's go for the simple solution
where we have a single syncobj bound to the context which we update to
point to the last job out_sync every time we submit a top-level job.

This approach implies reworking the way we create fences since we
need to capture the syncobj state at the time the fence is created.
Unfortunately, there's not SYNCOBJ_CLONE ioctl, which forces us to
export/create/import a fence so we have a new object that's not
subject to changes done to the context syncobj.

If we want to further optimize the logic, we should probably explore
some of those options:

1/ Adding array based SYNCOBJ ioctls (SYNCOBJ_{CREATE,DESTROY,CLONE}_ARRAY)
   so we can mitigate the cost of ioctls when we need to manipulate
   arrays of syncobjs
2/ Support synchronization jobs. That is, jobs that have a NULL job chain
   but an array of sync_in and a sync_out to allow creating
   synchronization points
3/ Add syncobj aggregators so we only have to wait on one syncobj from
   userspace. The syncobj aggregator would wait for all sub syncobjs to
   be signaled before signaling the top-level one.

Fixes: 64d6f56ad26f ("panfrost: Allocate syncobjs in panfrost_flush")
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7831>

panfrost: Make sure we always add a reader -> write dependency when needed

We shouldn't reset the ->writer field when a reader comes in because we
want subsequent readers to have a dependency on the writer too. Let's
add a new field encoding the last access type and use it to replace the
writer != NULL test.

Reported-by: Roman Elshin
Fixes: c6ebff3ecdde ("panfrost: Remove panfrost_bo_access type")
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7831>

st/mesa: enable compute shader derivatives in SPIR-V

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6799>

ac,radeonsi: implement GL_NV_compute_shader_derivatives

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6799>

winsys/amdgpu: use VRAM for command buffers if all VRAM is visible

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7951>

radeonsi: map PIPE_USAGE_STREAM to VRAM if all VRAM is visible

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7951>

radeonsi: unify uploaders and upload to VRAM if all VRAM is visible

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7951>

radeonsi: only use staging for linear textures when all VRAM is not visible

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7951>

radeons: only force staging uploads for VRAM when all VRAM is not visible

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7951>

ac: add radeon_info::all_vram_visible for Smart Access Memory

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7951>

radeonsi: initialize ctx and gfx_cs first, then allocators

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7951>

compiler: update gl_access_qualifier comments

Clarify ACCESS_NON_{READABLE,WRITEABLE} and update ACCESS_CAN_REORDER.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: infer writeonly

This isn't always done for GL because it could cause ImageAccess to be
incorrect: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3278

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: consider global stores

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: handle variable pointers

We might not be able to determine what variable a instruction accesses.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: add basic Vulkan support

This involves determining the variables referenced by intrinsics, setting and
using the access qualifier correctly and considering that images and buffers
can alias.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: rename can_reorder() and set ACCESS_NON_WRITEABLE in it

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: don't check restrict in can_reorder()

ACCESS_NON_WRITEABLE means that the memory is read-only, not the variable,
so we don't have to check for aliasing.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: check restrict before marking a variable as readonly

ACCESS_NON_WRITEABLE means that the memory is read-only, not the variable,
so we have to check for aliasing first.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: ignore barriers and coherent qualifier

It shouldn't matter whether an access/variable is coherent or not, just
that it's not written. The coherent qualifier doesn't mean anything with
read-only memory.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

nir/opt_access: don't ignore image arrays in process_variable()

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6483>

iris: update depth value for stages after fast clear depth

This should fix an outdated depth value visible from the shader side.

v1: (Nanley Chery)
- fix commit message

v2: (Nanley Chery)
- replace `fixes` tag by `mesa-stable`

Closes: #3883
Cc: 20.3 20.2 <mesa-stable>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Signed-off-by: Andrii Simiklit <andrii.simiklit@globallogic.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7838>

v3d: remove old tile blit code

This code was a direct copy from VC4, and it was disabled, as it did not
work.

Now that we have an implementation, let's remove it.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: implement tile-based blit operation

This implements fast-path blit using the TLB to blit from one buffer to
another, if conditions for allowing this are met.

v1:
 - Move checks in the code (Iago)

v2:
 - Use function to compute tile width and height (Iago)
 - Fix commit message (Iago)
 - Use surface size to compute draw_tiles_{x,y} (Iago)
 - Move checks (Iago)
 - Fix tile draw parameters (Iago)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: refactor set tile buffer size function

Change it to not require a v3d job to compute the tile buffer
parameters.

v1:
 - Pass nr_cbufs parameter to get tile buffer size function (Iago)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: implement tile buffer blits

This implements blit operation using the TLB.

It uses a source color buffer (bbuf) which will be blitted to color
buffer 0.

It also takes in account the number of samples for the input and output
so it can perform multisample resolve.

v1:
 - Fix comment (Iago)
 - Removed needless brackets (Iago)
 - Ensure msaa is correctly set (Iago)
 - Get rid of job->resolve (Iago)
 - Add rbuf as part of job's key (Iago)
 - Rename rbuf/rsurf by bbuf/bsurf (Iago)
 - Revert needless change (Iago)

v2:
 - Remove spurious change (Iago)
 - Add assert for safety reasons (Iago)
 - Add brackets in condition (Iago)
 - Fix commit message and title (Iago)
 - Do tile blit only for version >=4.0 (Iago)

v3:
 - Add assertion (Iago)
 - Fix comment (Iago)
 - Change commit title (Iago)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: add helper to check if format supports TLB resolve

It checks if the TLB can perform multisample resolve for the specified
format.

v1:
 - Fix commit title (Iago)

v2:
 - Fix identation (Iago)
 - Fix prototype style (Iago)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: split binning start from draw

Separate the binning part in the draw and make it public, so we can
invoke it later from a different part.

v1:
 - Ensure the job needs flush (Iago)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: store number of color buffers in job

Avoids requiring to calculate in other places.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: make set tile buffer size function public

This will be used later when implementing Tile-based blit function.

v1:
 - Fix commit title (Iago)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

v3d: force alpha to 1 when rendering RGBX formats

This adds an enum to the load tile buffer that forces the alpha channel
to be set to 1. This will be required later to load RGBX formats.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

util: function to check for rgbX format

Function that checks if the format has its alpha component forced to 1.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7816>

intel/tools/aubinator_error_decode: fix small memory leaks

Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7994>

intel/tools/aubinator_error_decode: cleanup path/file handling

Should help some compilers/static analyzers understand this code and avoid
things like this:

../src/intel/tools/aubinator_error_decode.c:850:19: warning: "path" may be used uninitialized in this function [-Wmaybe-uninitialized]
  850 |             ret = asprintf(&filename, "%s/%d/i915_error_state", path, minor);

Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7994>

r600/sfn: Initialize ShaderInputVarying members in constructors.

Fix defects reported by Coverity Scan.

uninit_member: Non-static class member m_interpolate is not initialized in this constructor nor in any functions that it calls.
uninit_member: Non-static class member m_lds_pos is not initialized in this constructor nor in any functions that it calls.
uninit_member: Non-static class member m_mask is not initialized in this constructor nor in any functions that it calls.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7909>

meson: Fix build with llvm-12.

This patch fixes this Meson build error.

$ meson builddir \
-Dshared-llvm=disabled
-Ddri-drivers=''
-Dbuild-tests=true \
-Dgallium-drivers=swrast \
-Dvulkan-drivers=''
[...]
/usr/bin/ld: src/gallium/auxiliary/libgallium.a(gallivm_lp_bld_misc.cpp.o): in function `llvm::InitializeNativeTarget()':
llvm/Support/TargetSelect.h:118: undefined reference to `LLVMInitializeX86TargetInfo'
/usr/bin/ld: llvm/Support/TargetSelect.h:119: undefined reference to `LLVMInitializeX86Target'
/usr/bin/ld: llvm/Support/TargetSelect.h:120: undefined reference to `LLVMInitializeX86TargetMC'
/usr/bin/ld: src/gallium/auxiliary/libgallium.a(gallivm_lp_bld_misc.cpp.o): in function `llvm::InitializeNativeTargetAsmPrinter()':
llvm/Support/TargetSelect.h:132: undefined reference to `LLVMInitializeX86AsmPrinter'
/usr/bin/ld: src/gallium/auxiliary/libgallium.a(gallivm_lp_bld_misc.cpp.o): in function `llvm::InitializeNativeTargetDisassembler()':
llvm/Support/TargetSelect.h:156: undefined reference to `LLVMInitializeX86Disassembler'

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7777>

ci/vc4: Skip VS dynamic loops tests that cause GPU hangs.

We've been getting spurious failures from the new VC4 CI, which I believe
are due to this set of tests (which have been showing up along with a GPU
hang report in the list of flaky tests in the failing jobs).  This was a
known issue I had in vc4.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7998>

radeonsi: fix the blit test for SW_64KB_R_X

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7908>

radeonsi: rename SI_TEST_DMA to SI_TEST_BLIT

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7908>

radeonsi: remove SDMA support

There are many issues with SDMA across many generations of hardware.
A recent example is that gfx10.3 suffers from random GPU hangs if
userspace uses SDMA.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7908>

radeonsi: rename buffer functions so as not to reference rings

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7908>

radeonsi: move si_screen_clear_buffer into si_compute_blit.c w/o SDMA option

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7908>

ci/freedreno: Detect the cheza power management bus error and restart.

This is an issue on the cheza platform, the theory is due to some old
firmware bug that will be fixed in future platforms.  Given that cheza was
a target that didn't get released and we expect future platforms to be
fixed, just detect the issue and restart.

I've noticed this error in my CI monitoring less than once a week.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7993>

android: spirv: fix '::' typo in gen rules

among all Android gen rules '::' was used only here to declare dependencies;
mesa development and stable branch are worth receiving the fix

Fixes the following building errors with Android 7:

obj/STATIC_LIBRARIES/libmesa_nir_intermediates/spirv/gl_spirv.P:184: *** target file
gen/STATIC_LIBRARIES/libmesa_nir_intermediates/spirv/vtn_generator_ids.h' has both : and :: entries.  Stop.

Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Fixes: 1070bba19e26 ("android: fix SPIR-V -> NIR build")
Reported-by: youling257 <youling257@gmail.com>

mesa/st: set reserved storage for params+values to 16

zink apparently has a lot of uniform data, and 8x8 isn't enough to avoid realloc

fixes mesa/mesa#3930

Fixes: eda37fb2 ("mesa: properly disallow param list reallocation")

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7990>

meson: fix multiline string warning

gets rid of: WARNING: Newline character in a string detected, use '''
(three single quotes) for multiline strings instead.

Reviewed-by: Yevhenii Kharchenko <yevhenii.kharchenko@globallogic.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7946>

util: Add cnd_monotonic to Makefile.sources

Fixes: 33a78948283 ("util,radv: Cross-platform monotonic condition variable")
Acked-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7980>

mesa: remove MAX_3D_TEXTURE_LEVELS, MAX_CUBE_TEXTURE_LEVELS

they are redundant with MAX_TEXTURE_LEVELS

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7956>

mesa: remove code for old (mostly unsupported) GL_NV_point_sprite

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7956>

mesa: fix glPopAttrib for GL_COORD_REPLACE for r200

Fixes: 959380dc "mesa: more optimizations in glPopAttrib (colormask, drawbuffers, coord replace)"

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7956>

turnip: always set LRZ registers to zero for 3d clear/blit

Apparently LRZ will be read/written regardless of depth being enabled or
not, so we have to make sure these registers are zero.

Fixes: 1d83f5ae8435 ("turnip: disable LRZ on vkCmdClearattachments() 3D fallback path")

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>

turnip: move up LRZ invalidate in CmdClearAttachments

There is an early return if cmd->state.predication_active is true, so do
the LRZ invalidate before that.

Fixes: 2f79e0066405 ("turnip: disable LRZ on vkCmdClearAttachments()")

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>

turnip: do not emit draw states in draw_cs outside of renderpass

This avoids a possible issue with MSAA sysmem clears, which use a 3D clear
path which assumes draw states are disabled, and are emitted in draw_cs in
BeginRenderPass.

(checking for TU_CMD_DIRTY_DRAW_STATE also allows not emitting the draw
states if they will be re-emitted on the next draw anyway. the previous
patch makes it so TU_CMD_DIRTY_DRAW_STATE is always set outside of
renderpasses)

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>

turnip: correctly disable draw states outside of renderpasses

* do the disable in EndRenderPass2 to fix the missing disable for sysmem
* we don't need a disable at the end of every tile, or between binning pass
  and gmem pass (the first draw in draw_cs emits all the draw states)

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>

turnip: always emit LRZ draw state in DIRTY_DRAW_STATE path

The packet size is constant and assumes all states, except for the 2 input
attachment states. (this means we get an invalid packet if DIRTY_LRZ isn't
set when DIRTY_DRAW_STATE is set).

Fixes: 3c07a1499863 ("turnip: enable LRZ")

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>

turnip: do not include compute stage in pipeline_builder

This avoids emitting compute-related state in the graphics pipeline
(tu6_emit_xs_config was being called for compute stage).

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>

turnip: no linear_to_srgb for alpha channel for gmem clear value packing

Alpha channel is always linear (oops).

Fixes: ddac5933f8f3 ("turnip: call packing functions directly for pack_gmem_clear_value")

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>

gallium/u_threaded: optimize set_constant_buffer

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

gallium/u_threaded: don't make a local copy of pipe_draw_start_count

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

gallium/u_threaded: don't copy the indexbuf pointer if we overwrite it

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

gallium/u_threaded: set has_user_indices = false in the driver thread

to remove some overhead from the main thread.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

gallium/u_threaded: don't pass index bounds to the driver to decrease overhead

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

glthread: count batch space in units of uint64_t elements

This removes one x86 shr instruction from _mesa_glthread_allocate_command.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

glthread: change sizes to unsigned or size_t where needed

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

glthread: use uint64_t to declare the batch buffer instead of align(8)

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

glthread: use glthread->used instead of glthread->next_batch->used

remove one pointer dereference by having "used" in glthread too

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7719>

zink: use shader keys for samplemask

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7801>

zink: change a memcmp==0 to !memcmp

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7801>

zink: initial shader key implementation

this is a squashed version of a previously-reviewed series
ref https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7193

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7801>

vulkan/device_select: Store Vulkan vendorID and deviceID as uint32_t

Vulkan uses 32-bit IDs, compared to PCI/USB 16-bit ones.

Some driver vendorIDs do exceed 0xffff, including MESA (used by lavapipe).

Without this, the value will be truncated, and device select layer
will not match expected (or any) device.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3935
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7977>

gallium/aux: Update scons build for u_tracepoints.[ch]

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7974>

gallium/aux: Split u_tracepoints.[ch] generation

Allow for separate rules to be used to generate the .c and .h (partially
to make it easier for scons build, and partially because that is just a
better practice).

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7974>

gallium/aux: Avoid creating queue when traces not enabled

Avoids unnecessarily creating an extra thread when tracing is not
enabled.  Which also side-steps a chrome/ium sandbox issue with
the setscheduler call in u_queue.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7974>

nir: improve divergence analysis for loads with non-uniform resources

If ACCESS_NON_UNIFORM is not specified, we can assume the resource is
uniform. This requires nir_lower_non_uniform_access to remove that flag.

A few Detroit: Become Human shaders use a index sourced from a fragment
input which is expected to be uniform.

shader-db (Navi):
Totals from 8 (0.01% of 127638) affected shaders:
SGPRs: 224 -> 384 (+71.43%)
VGPRs: 208 -> 112 (-46.15%)
CodeSize: 5360 -> 5344 (-0.30%); split: -1.49%, +1.19%
Instrs: 1036 -> 1028 (-0.77%); split: -1.93%, +1.16%
VMEM: 1320 -> 608 (-53.94%)
SMEM: 384 -> 336 (-12.50%); split: +14.58%, -27.08%
VClause: 24 -> 16 (-33.33%)
SClause: 48 -> 56 (+16.67%)
PreSGPRs: 124 -> 216 (+74.19%)
PreVGPRs: 168 -> 88 (-47.62%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5201>

nir/lower_non_uniform: remove non_uniform flags after lowering

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5201>

iris: fix memleak for query_buffer_uploader

In the Chrome WebGL Aquarium  stress test, 20 instances of Chrome will run
Aquarium  simultaneously over 20+ hours. That causes Chrome crash.
During the stress, glBeginQueryIndexed is called frequently.

1.Each query will only use 32 bytes from query_buffer_uploader. After the offset
exceed 4096, it will alloc new buffer for query_buffer_uploader->buffer
and release the old buffer.

2.But iris_begin_query will call u_upload_alloc when the offset changed, and it
will increase the query_buffer_uploader->buffer->reference.count every time
when it called u_upload_alloc.

3.So when u_upload_release_buffer try to release the resource of
query_buffer_uploader->buffer, its reference.count is
already equal to 129. pipe_reference_described will only decrease its reference
count to 128.So it never called old_dst->screen->resource_destroy.

4.The old resouce bo will never be freeed. And chrome will called mmap every time
when it alloc new resource bo.

5. Chrome process map too many vmas in its process. Its map count exceed the
sysctl_max_map_count which is 65530 defined in kernel.

6. When iris_begin_query want to alloc new resource bo, it will meet NULL pointer
because mmap return failed. Finally chrome crashed when it access this NULL resource
bo.

The fix is decrease the reference count in iris_destroy_query.

Patch is verified by chrome webgl Aquarium test case for more than 72 hours.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Yang Shi <yang.a.shi@intel.com>
Reviewed-by: Alex Zuo <alex.zuo@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7890>

aco: use UINT64_C on 64 bit constant arguments

avoids errors seen when building on OpenBSD/amd64

../src/amd/compiler/aco_instruction_selection.cpp:1677:62: error: ambiguous conversion for functional-style cast from 'unsigned long' to 'aco::Operand'
            bld.vop3(aco_opcode::v_mul_f64, Definition(dst), Operand(0x3FF0000000000000lu), tmp);
                                                             ^~~~~~~~~~~~~~~~~~~~~~~~~~~
glibc uses unsigned long for uint64_t on LP64 archs and unsigned long long for
uint64_t on ILP32 archs.  On OpenBSD unsigned long long is used for uint64_t
on all archs.

The Operand constructors are uint8_t uint16_t uint32_t uint64_t
use UINT64_C so lu or llu suffix will be used as needed.

Fixes: df645fa369d ("aco: implement VK_KHR_shader_float_controls")
Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7944>

mesa/st: lower 64 bit ops to scalar before lowering to soft-float

The fp64 emulation is not prepared for vectorized 64 bit code, so
if the driver doesn't ask for lowering to scalar by itself, do it before
lowering to soft-fb, and run a vectorizazion pass afterwards.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7961>

radv: disable alphaToOne feature

The feature was exposed but completely ignored by the driver. Other
AMD drivers don't expose it as well, probably because it's complicated
to implement alpha-to-coverage properly. Let's disable it.

Cc: mesa-stable.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7966>

driconf: add allow_incorrect_primitive_id option

And enable it for SPECviewperf.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7078>

vbo/dlist: only use merged primitives when it's ok to do so

Merging primitives generates incorrect gl_PrimitiveID[In] values.
So make merged primitives construction non-destructive and fallback
to drawing with original primitives if a program reads gl_PrimitiveId.

This commit adds _mesa_update_primitive_id_is_unused modeled after
_mesa_update_allow_draw_out_of_order to update ctx->_PrimitiveIDIsUnused
each time shaders are updated.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7078>