etnaviv: Emit PE.ALPHA_COLOR_EXT* on GPUs with half-float support

At least GC880 (iMX6S), GC2000 (iMX6Q) blobs do not emit the
PE.ALPHA_COLOR_EXT0 and PE.ALPHA_COLOR_EXT1 into the command
stream. The GCnano (STM32MP1) is not affected by this change
either. This is because neither of these GPUs support the
half-float feature.

Emit PE.ALPHA_COLOR_EXT* in etnaviv only if half-float support
is present in the GPU. This fixes all of the currently failing
dEQPs in this group:
  dEQP-GLES2.functional.fragment_ops.blend.*

Fixes: 76adf041f25 ("etnaviv: fix blend color on newer GPUs")
Signed-off-by: Marek Vasut <marex@denx.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4277>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4277>

gallivm: disable rgtc/latc SNORM accellerated fetches

Unfortunately this appears to be bugged (it seems the piglit tests aren't
quite exhaustive enough). I'm almost certain it's the lerp
(lp_build_lerpdxta()) which doesn't handle signed numbers correctly, let's
disable for now.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4311>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4311>

rbug: do not return void-value

Returning a void-value is nonsensical, and in this case it seems like a
mistake.

This eliminates a warning when building on MSVC.

Fixes: fb04e5da97d ("gallium: add pipe_screen::finalize_nir")
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4297>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4297>

rbug: clean up cast-warnings

Similarly to the previous cast; on 64-bit Windows, unsigned long is
32-bit, and casting a pointer to a non-matchin bit-width integer produce
warnings. So let's use uintpre_t for this purpose instead.

Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4297>

pipebuffer: clean up cast-warnings

This code produces warnings, so let's fix that. The problem is that
casting a pointer to an integer of non-pointer-size triggers warnings on
MSVC, and on 64-bit Windows unsigned long is 32-bit large.

So let's instead use uintptr_t, which is exactly for these kinds of
things.

While we're at it, let's make the resulting index a plain "unsigned",
which is the type this originated from before we started with this
cast-dance.

Fixes: 1a66ead1c75 ("pipebuffer, winsys/svga: Add functionality to update pb_validate_entry flags")
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4297>

vulkan/overlay: Add a workaround semaphore for application presenting without one

When an application calls vkQueuePresent() on a different queue than
the one we run our drawing on and it doesn't give a semaphore to wait
on, let's insert our own semaphore so that we don't race the
application's drawing.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2540
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3893>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3893>

ac: fix ac_build_is_helper_invocation when postponed_kill is null

If there was no demote() in the shader, ac_build_is_helper_invocation
behaves exactly the same as ac_build_load_helper_invocation, i.e.
the helper lanes are the same as they were at the beginning of the shader.

Fixes: de57ea2a3da ("amd/llvm: implement nir_intrinsic_demote(_if) and nir_intrinsic_is_helper_invocation")
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4301>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4301>

nir: update uses_demote flag in discard_to_demote pass

Otherwise the ctx.ac.postponed_kill will not be allocated.

Fixes: ce87da71e93 ("nir: add pass to lower discard() to demote()")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2662
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4301>

glsl/lower_precision: Lower builtins depending on arguments

When an ir_call is encountered that invokes a builtin, it will now try
to generate a lowered version of the builtin. This only happens if all
of the arguments to the function are lowerable. Previously the builtin
would be inlined before the lowering pass is invoked and then the
implementation would be lowered as a consequence of the pass. However
this causes problems if the builtin has multiple arguments and the
implementation has operations on only a few of the arguments before
combining it with the others. In that case the entire builtin should
only be lowered if all of the arguments are lower precision. The
previous approach would end up lowering only parts of the
implementation.

The lowered implementations are cached in a hash table in case they can
be reused.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

glsl: Inline builtins in a separate pass

Previously, the ir_call functions for builtin functions were replaced
with the inline implementation immediately after being added to the
instruction list. This patch replaces that with a separate pass that
lowers them after the conversion from AST to IR is complete. This will
be useful to be able to insert some handling for the precision lowering
pass before the inlining. This needs to happen because the precision
of the operations in the inlined implementation depends on the highest
precision of all of the arguments to the call.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

freedreno/ir3: enable nir_opt_loop_unroll on a6xx

If precision lowering happens at GLSL IR, loop_analysis at IR doesn't
work as expected since it can't handle things like:

"(expression bool < (expression float16_t f2fmp (var_ref ndx) ) (constant float16_t (1.000000)) )"

So we'd rather do this optimization at the NIR stage.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

freedreno/ir3: Lower bools to bitsize

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

nir: add a bool bitsize lowering pass

The pass lowers 1-bit booleans produced by NIR to the native bitsize
of the operations that produce them.

v2: change on lower_load_const_instr after upstream changes. Added
    TODO2 to explain it, as it was not properly tested yet (see
    already existing TODO) (Neil)

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

freedreno: Enable mediump lowering

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

glsl: Add unit tests for the lower_precision pass

Adds a unit tests script that invokes the standalone compiler with
--lower-precision and verifies that lowered operations are being used.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

glsl/standalone: Add an option to lower the precision

Adds a --lower-precision option that just sets the LowerPrecision
compiler option. That way it can be used in unit tests to test the
precision lowering pass.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

glsl: Add an IR lowering pass to convert mediump operations to 16-bit

This works by finding the first rvalue that it can lower using an
ir_rvalue_visitor. In that case it adds a conversion to float16
after each rvalue and a conversion back to float before storing
the assignment.

Also it uses a set to keep track of rvalues that have been
lowred already. The handle_rvalue method of the rvalue visitor doesn’t
provide any way to stop iteration. If we handle a value in
find_precision_visitor we want to be able to stop it from descending into
the lowered rvalue again.

Additionally this pass disallows converting nodes containing non-float.
The can_lower_rvalue function explicitly excludes any branches
that have non-float types except bools. This avoids the need to have
special handling for functions that convert to int or double.

Co-authored-by: Hyunjun Ko <zzoon@igalia.com>
v2. Adds lowering for texture samples

v3. Instead of checking whether each node can be lowered while walking the
tree, a separate tree walk is now done to check all of the nodes in a
single pass. The lowerable nodes are added to a set which is checked
during find_precision_visitor instead of calling can_lower_rvalue.

v4. Move the special case for temporaries to find_lowerable_rvalues. This
needs to be handled while checking for lowerable rvalues so that any
later dereferences of the variable will see the right precision.

v5. Add an override to visit ir_call instructions and apply the same
technique to override the precision of the temporary variable in the
same way as done for builtin temporaries and ir_assignment calls.

v6. Changes the pass so that it doesn’t need to lower an entire subtree in
order do perform a lowering. Instead, certain instructions can be
marked as being indepedent of their child instructions. For example,
this is the case with array dereferences. The precision of the array
index doesn’t have any bearing on whether things using the result of
the array deref can be lowered.

Now, only toplevel lowerable nodes are added to the lowerable_rvalues
instead instead of additionally adding all of the subnodes.

It now also only needs one hash table instead of two.

v7. Don’t try to lower sampler types. Instead, the sample instruction is
now treated as an independent point where the result of the sample can
be used in a lowered section. The precision of the sampler type
determines the precision of the sample instruction. This also means
the coordinates to the sampler can be lowered.

v8. Use f2fmp instead of f2f16.

v9.  Disable lowering derivatives calcualtions, which might not work
properly on some hw backends.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

glsl/hierarchical_visitor: Call leave_callback on leaf nodes

Previously for leaf ir_instructions only the enter callback was
called. This makes it a bit difficult to make a pass that wants to
visit every instruction using a stack. Making it call the leave
callback as well makes it behave less surprisingly.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

glsl: Add a method to get precision from a deref instruction

Adds ir_dereference::precision(). For a normal variable dereference,
the precision comes from the variable. For a record member it comes
from the field within the record. For an array it can come from
either, depending on where the underlying array is stored. The method
recursively walks the derefs until it finds one of the first two.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3885>

i965/iris: fix crash when calling GetPerfQueryDataINTEL

On a query that was never begun.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4302>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4302>

glthread: compile marshal_generated.c faster by breaking it up into 8 files

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4270>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4270>

glthread: declare marshal and unmarshal functions as non-static

Declare them in the header file. Then we can split marshal_generated.c
into multiple files.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4270>

glthread: inline SET_func and add -O1 to build _mesa_create_marshal_table faster

The compile time of marshal_generated.c improved from 30.1s to 12.4s.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4270>

radv: enable VK_KHR_8bit_storage on GFX6-GFX7

Enabling a Vulkan extension doesn't mean that all features need
to be implemented. DOOM Eternal crashes at launch if that ext
is not supported but it doesn't matter if the features are enabled
or not.

Let's enable it like we did for VK_KHR_16bit_storage.

Cc: 19.3 20.0 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4299>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4299>

util/u_process: fix Windows build

Reported by Brian Paul.

Fixes: f8f1413070a ("util/u_process: add util_get_process_exec_path")
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4303>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4303>

pan/bi: Rewrite aligned vectors as well

This still isn't optimal, but it handles another common case where we
have a vector "prefix" and can rewrite directly. The last case is one
where writemasks and such would start coming into play and - for the
moment - is not worth the hike in complexity.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4288>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4288>

pan/bi: Lower combines to rewrites for scalars

This avoids unneeded moves for scalars. It still generates suboptimal
code for vectors, however.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4288>

pan/bi: Ingest vecN directly (again)

Last time, I swear. We still generate writemasks but SSA-like ones and
do the lowering ourselves.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4288>

turnip: implement depth clamp

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4293>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4293>

turnip: fix znear clipping

Vulkan clips znear at 0 instead of -1.

Fixes dEQP-VK.draw.inverted_depth_ranges.nodepthclamp_*

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4293>

freedreno/registers: more GRAS_CL_CNTL bits, Z_CLAMP

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4293>

aco: implement 64-bit VGPR constant copies in handle_operands()

64-bit VGPR constant copies can happen because of 64-bit constant copy
propagation. Since this optimization is beneficial and more annoying to
deal with in the optimizer, I've implemented 64-bit VGPR constant copies
in handle_operands().

This also sets copy_operation::size correctly for 64-bit constant copies.

Cc: 20.0 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4260>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4260>

aco: remove dead code in handle_operands()

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/4260>

nir/gather_info: fix per-vertex handling in try_mask_partial_io

pipeline-db (Navi, ACO):
Totals from affected shaders:
SGPRS: 6432 -> 6432 (0.00 %)
VGPRS: 11924 -> 11924 (0.00 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Scratch size: 1596 -> 1596 (0.00 %) dwords per thread
Code Size: 575524 -> 518620 (-9.89 %) bytes
LDS: 12187 -> 12187 (0.00 %) blocks
Max Waves: 2695 -> 2695 (0.00 %)

Helps a few hundred Dark Souls 3 shaders.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4190>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4190>

radeonsi: enable workarounds for YoYo engine based games

Without the radeonsi_sync_compile option the games crashes at
startup.
The engine seems to be using a custom global new operator and
it doesn't plays well with multithreading it seems.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1310
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1271
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1272
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1288
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2611
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4181>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4181>

util/xmlconfig: add new sha1 application attribute

This is useful to enable workarounds for applications with a generic name.

For instance all games made with the YoYo game engine have the same executable
name "runner".

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4181>

util/u_process: add util_get_process_exec_path

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4181>

util/os_file: extend os_read_file to return the file size

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4181>

radeonsi: clarify the conditions when FLUSH_AND_INV_DB is needed

FLUSH_AND_INV_DB should be done when we're changing surface state
registers of a bound depth target.

When depth_clear_value changes, si_state will modify
S_028038_ZRANGE_PRECISION so we need to flush the DB caches.

Verified with the captures from bugs cited below.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1283
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1330
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4263>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4263>

intel/dump_gpu: Handle a bunch of getparam in the no-HW case

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4250>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4250>

intel/dump_gpu: Add an ensure_device_info helper

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4250>

anv: Stop fetching the timestamp frequency ourselves

gen_get_device_info_from_fd fetches the timestamp frequency from the
kernel.  ANV also carrying code for it is redundant.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4250>

egl/android: enable/disable KHR_partial_update correctly

Commit f3728816af (egl/android: require ANDROID_native_fence_sync
for buffer age) re-added some stale code removed in commit
b4345da8762 (egl/android: Delete set_damage_region from egl dri
vtbl).  Remove it now.

Commit b4345da8762 assumes KHR_partial_update is only
driver-dependent.  That is mostly true except that the extension
also introduces buffer age query, which depends on
ANDROID_native_fence_sync on Android.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Lepton Wu <lepton@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4235>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4235>

ci: Ban the recent popular freedreno a630 intermittent failure.

This popped up last thursday.  The only relevant code commit was my pixel
center half integer change, but the more likely thing to me seems to be
having shuffled the test order by introducing more skips the day before.

Link: https://gitlab.freedesktop.org/mesa/mesa/issues/2670
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4287>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4287>

st/mesa: fix use of uninitialized memory due to st_nir_lower_builtin

reported by valgrind

Cc: 19.3 20.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4274>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4274>

aco: fix boolean undef regclass

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4285>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4285>

lima: Add missing source file to Android.mk

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Roman Stratiienko <roman.stratiienko@nure.ua>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4283>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4283>

intel/tools/aubinator_error_decode: Decode ring buffers from HEAD to TAIL

Capture the HEAD and TAIL register values from the dump and
properly index the ring buffer using those. Previously we would
decode the ring buffer from the beginning, printing out whatever
happened to be there.

Also, properly pass the `from_ring` parameter to gen_print_batch()
so that decoding doesn't stop once MI_BATCH_BUFFER_END is
encoutered.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4261>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4261>

docs/features: Update virgl OpenGL 4.5 features
GL_ARB_clip_control and GL_KHR_robustness are now expose in the guest.

Signed-off-by: Elie Tournier <elie.tournier@collabora.com>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4160>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4160>

intel/tools/aubinator_error_decode: read HW Context before other batches

The hardware context buffer has state that was set before the
batch started. By decoding it first, references to things like
Dynamic State Base Address are decodable in the command batches.

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4246>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4246>

iris: Set patch count threshold in 3DSTATE_HS

Lets specifiy maximum number of patches that will be accumulated before
a thread is dispatched.

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3563>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3563>

anv: Set patch count threshold in 3DSTATE_HS

Lets specifiy maximum number of patches that will be accumulated before
a thread is dispatched.

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3563>

intel/compiler: Track patch count threshold

Return the number of patches to accumulate before an 8_PATCH mode thread
is launched.

v2: (Kenneth Graunke)
- Track patch count threshold instead of input control points.

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3563>

intel/genxml: Add patch count threshold field on gen12

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3563>

gitlab-ci/traces: Add Vulkan sample entries for POLARIS10

v2:
  - Updated commit log.

Signed-off-by: Andres Gomez <agomez@igalia.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4103>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4103>

gitlab: add bug report template

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4089>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4089>

aco: emit IR in IF's merge block instead if the other side ends in a jump

Fixes NIR such as:
if (divergent) {
   a = sgpr()
} else {
   break;
}
use(a)

Previously we would have emitted:
if (divergent) {
   a = sgpr()
}
if (!divergent) {
   break;
}
use(a)

But "a" isn't available at it's use. Now we emit:
if (divergent) {
}
if (!divergent) {
   break;
}
a = sgpr()
use(a)

pipeline-db (Navi):
Totals from affected shaders:
SGPRS: 1936 -> 1936 (0.00 %)
VGPRS: 1264 -> 1264 (0.00 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 159408 -> 159152 (-0.16 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 81 -> 81 (0.00 %)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
CC: <mesa-stable@lists.freedesktop.org>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2557
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3658>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3658>

aco: improve check for unreachable loop continue blocks

The old code would have previously caught:
loop {
   ...
   break
}
when it was meant to just catch:
loop {
   if (...)
      break
   else
      break
}

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3658>

aco: skip NIR in unreachable merge blocks

NIR removes most of this but undef instructions for loop header phis can
remain. These were harmless because ACO would DCE them itself.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3658>

aco: handle when ACO adds new continue edges

Usually a loop ends with a uniform continue. If it doesn't and we end up
adding our own continue edges (because of continue_or_break or divergent
breaks at the end), we have to add extra operands to the loop header phis.

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/3658>

aco: handle missing second predecessors at merge block phis

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3658>

aco: set has_divergent_branch for discards in loops

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3658>

gitlab-ci: add python3-requests to the test-vk container

After 90a39af5f65 ("ci: Drop the git dependency in tracie"), we have
this error in the radv-polaris10-traces job:

"
...

+ /builds/tanty/mesa/artifacts/tracie/tests/test.sh
tracie_succeeds_if_all_images_match: Fail
Traceback (most recent call last):
  File "/tmp/tracie.test.glY0O23HJo/tracie.py", line 6, in <module>
    import requests
ModuleNotFoundError: No module named 'requests'

...
"

v2:
  - Updated commit log to be more descriptive (Michel).

Fixes: 90a39af5f65 ("ci: Drop the git dependency in tracie")
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4237>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4237>

radv/llvm: fix subgroup shuffle for chips without bpermute

bpermute only exists on GFX8+ and only with Wave32 on GFX10. Instead
we have to use readlane with a waterfall loop to defeat the LLVM
backend.

This fixes DOOM Eternal which requires subgroup shuffle.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4284>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4284>

panfrost: Align Android makefiles with recent changes

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Signed-off-by: Roman Stratiienko <roman.stratiienko@nure.ua>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4280>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4280>

gitlab-ci: add a bunch of new fossils from the Sascha Vulkan demos

The whole fossils-db is only 448KB of data which is pretty small.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4082>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4082>

gitlab-ci: add a new stage for RADV CI

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4082>

gitlab-ci: compile fossils with more ASICs

I think we want to cover these 3 generations at the barely minimum.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4082>

gitlab-ci: compile fossils with both RADV compiler backends (LLVM/ACO)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4082>

gallium/gallivm: Remove workaround disabling AVX code for newer CPUs

The change enables using full 256-bit AVX and AVX2 instructions
on newer platforms.

Reviewed-by: Alok Hota <alok.hota@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4225>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4225>

radv/winsys: spoof some values for num_render_backends in the null winsys

To avoid crashes when RADV_FORCE_FAMILY is set to GFX9+ because
num_render_backends is used to compute binning state.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4282>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4282>

radv/winsys: fix wrong PCI ID for Vega10 in the null winsys

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/4282>

glsl: Restore the IsES flag on the shader when reading from cache.

I found that when trying to MESA_SHADER_CAPTURE_PATH a trace, I was
getting "GLSL >= 3.00" for the ES shaders I was trying to capture.
Keeping this metadata in the cached shader program lets us capture
correctly.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4219>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4219>

gallivm: add support for rgtc/latc fetches.

Annoyingly heaven uses rgtc2 snorm but this at least avoids
the function call overheads to the util fetch functions.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3924>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3924>

gallivm/s3tc: split out dxt5 alpha code

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3924>

intel: Add TGL PCI ID

Ref: Bspec 44455
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

intel: Update TGL PCI strings

Ref: Bspec 44455
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

pan/bi: Pack csel4 opcodes

These are pretty straightforward but there's a lot of details to keep
straight. In the IR, we keep a general logical comparator and types
separately; in the hardware, the type gets fused with a (much more)
limited number of comparators. So there's a fair bit of code here to
account for these differences, fusing in the type information, and
changing up argument order as necessary to make it actually correct.
Anything to save a bit!

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Default csel to "!= 0" mode

This way we always have regular csel conditions instead of a weird
.always special case for 3-src CSEL mode.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Use bi_lookup_immediate when packing

This gets us part of the way there to packing lo/hi separately. A little
more work is needed to do this "properly", but hey.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Respect shift when printing immediates

We allow packing multiple immediates in, but we were missing this in the
print.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Implement csel fusing

When generating csel instructions, we can peak to see what condition is
being used. If we're using a "nice" condition, we can fuse it in with
the csel itself, ideally letting the condition itself be DCE'd away.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Add `soft` NIR->BIR condition translation

We would like to use this routine opportunistically when fusing
conditions into csels and branches, so let's add a mode where we don't
abort.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Remove hacks for 1-bit booleans in IR

Now that we lower them away, a bunch of special cases disappear.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Lower bool to ints

Currently we lower to int32, but once mediump lands we'll be ready for
that too.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Pack LD_ATTR

Also requires the usual R61/62 games.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Pack st_vary

This should let varying writes go through finally.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Add store_channels property

It can't be inferred from the usual writemask since stores don't write
to a register destination.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Generalize data register setting

So we can use it for stores too.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Flesh out st_vary IR

We need to make the semantics of BI_VECTOR a bit more precise -
vectorize only the first argument, not all of them. This is enough for
current and future users, as far as I know.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Pack ld_var_addr

Choo choo.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Pack ld_ubo ops

Routes some infrastructure to do so at least slightly generically but
we'll see.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Add bi_load32_components helper

Pattern seems to crop up a lot.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Include UBO index for sysval reads

Trivially zero.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Index out constants in instructions

We rewrite BIR_INDEX_CONSTANT (and _ZERO) to preassigned constant ports
when assign uniform_const for the bundle. There are a lot of issues
raised here, unfortunately, and the implementation here is woefully
incomplete with a nasty hack for loads... nevertheless, it's somewhere
to start.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Document constant related errata(?)

We're not totally sure what's up with this but Connor says if you
violate it Bad Things happen in your shader. I think this might be an
issue affecting early Bifrost (G71, ..?); when we know more we can look
into patching in a fix.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Pack a constant quadword

The piping isn't there to make use of it yet, but this stubs out
constant support at the clause level.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Add move lowering pass

We need ALU mostly scalarized, but we get vector moves created from
lower_vec_to_mov so let's scalarize that ourselves rather than bother
NIR.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Add bi_emit_before helper

For BIR lowering passes.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>

pan/bi: Implement FMA/MOV without modifiers

We split off MOV from FMOV since the canonical move on Bifrost doesn't
accept modifiers. (We can still do fmov, but with something like add-0.)
This will also make copyprop a little nicer, I think. Anyway, the
non-modifier version we can implement as-is for FMA.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4276>