glsl: fix constant packing for 64-bit big endian.

In a piglit run on s390 a lot of double tests fail, explicitly
packing/shifting things rather than using memcpy seems to help

Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5679>

docs: add a page explaining the GitLab CI and the Intel CI

This explains what they are, what they do and how to use them.

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Clayton Craft <clayton.a.craft@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2250>

mesa/swrast: use logf2 instead of util_fast_log2

The fast version is apparently not accurate enough. I wrote a very
simply test program that called logf2 and the old LOG2 function 100000
times. Across that the two functions had very similar run times, neither
appeared meaningfully faster, so the optimization of bringing back yet
another way to calculate log2f seems pointless.

Fixes: bd4e769515345a6b20562310334bc828c0bb6605
       ("replace LOG2 with util_fast_log2")

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2856
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5406>

anv: disable i915_perf warning on non-Linux

$ vkcube
INTEL-MESA: warning: Performance support disabled, consider sysctl dev.i915.perf_stream_paranoid=0

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5461>

meson: Revert commit overriding C++ standard with gnu++11 on ppc64el

Since a few versions, mesa now needs c++14 and compiling with gnu++11
on ppc64el fails.
Let's use the default standard and fix the collision of types between
c++ and altivec in a another patch.

Cc: mesa-stable
Signed-off-by: Frédéric Bonnard <frediz@debian.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4948>

clover: Fix types collision between c++ and altivec

For that, we undefine bool, vector, pixel as advised by altivec.h in the
specific case that defines them.

Cc: mesa-stable
Signed-off-by: Frédéric Bonnard <frediz@debian.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4948>

nir: Propagate *2*16 conversions into vectors

If we have code like:

   ('f2f16', ('vec2', ('f2f32', 'a@16'), '#b@32'))

We would like to eliminate the conversions, but the existing rules can't
see into the the (heterogenous) vector. So instead of trying to
eliminate in one pass, we add opts to propagate the f2f16 into the
vector. Even if nothing further happens, this is often a win since then
the created vector is smaller (half2 instead of float2). Hence the above
gets transformed to

   ('vec2', ('f2f16', ('f2f32', 'a@16')), ('f2f16', '#b@32'))

Then the existing f2f16(f2f32) rule will kick in for the first component
and constant folding will for the second and we'll be left with

   ('vec2', 'a@16', '#b@16')

...eliminating all conversions.

v2: Predicate on !options->vectorize_vec2_16bit. As discussed, this
optimization helps greatly on true vector architectures (like Midgard)
but wreaks havoc on more modern SIMD-within-a-register architectures
(like Bifrost and modern AMD). So let's predicate on that.

v3: Extend for integers as well and add a comment explaining the
transforms.

Results on Midgard (unfortunately a true SIMD architecture):

total instructions in shared programs: 51359 -> 50963 (-0.77%)
instructions in affected programs: 4523 -> 4127 (-8.76%)
helped: 53
HURT: 0
helped stats (abs) min: 1 max: 86 x̄: 7.47 x̃: 6
helped stats (rel) min: 1.71% max: 28.00% x̄: 9.66% x̃: 7.34%
95% mean confidence interval for instructions value: -10.58 -4.36
95% mean confidence interval for instructions %-change: -11.45% -7.88%
Instructions are helped.

total bundles in shared programs: 25825 -> 25670 (-0.60%)
bundles in affected programs: 2057 -> 1902 (-7.54%)
helped: 53
HURT: 0
helped stats (abs) min: 1 max: 26 x̄: 2.92 x̃: 2
helped stats (rel) min: 2.86% max: 30.00% x̄: 8.64% x̃: 8.33%
95% mean confidence interval for bundles value: -3.93 -1.92
95% mean confidence interval for bundles %-change: -10.69% -6.59%
Bundles are helped.

total quadwords in shared programs: 41359 -> 41055 (-0.74%)
quadwords in affected programs: 3801 -> 3497 (-8.00%)
helped: 57
HURT: 0
helped stats (abs) min: 1 max: 57 x̄: 5.33 x̃: 4
helped stats (rel) min: 1.92% max: 21.05% x̄: 8.22% x̃: 6.67%
95% mean confidence interval for quadwords value: -7.35 -3.32
95% mean confidence interval for quadwords %-change: -9.54% -6.90%
Quadwords are helped.

total registers in shared programs: 3849 -> 3807 (-1.09%)
registers in affected programs: 167 -> 125 (-25.15%)
helped: 32
HURT: 1
helped stats (abs) min: 1 max: 3 x̄: 1.34 x̃: 1
helped stats (rel) min: 20.00% max: 50.00% x̄: 26.35% x̃: 20.00%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 16.67% max: 16.67% x̄: 16.67% x̃: 16.67%
95% mean confidence interval for registers value: -1.54 -1.00
95% mean confidence interval for registers %-change: -29.41% -20.69%
Registers are helped.

total threads in shared programs: 2471 -> 2520 (1.98%)
threads in affected programs: 49 -> 98 (100.00%)
helped: 25
HURT: 0
helped stats (abs) min: 1 max: 2 x̄: 1.96 x̃: 2
helped stats (rel) min: 100.00% max: 100.00% x̄: 100.00% x̃: 100.00%
95% mean confidence interval for threads value: 1.88 2.04
95% mean confidence interval for threads %-change: 100.00% 100.00%
Threads are [helped].

total spills in shared programs: 168 -> 168 (0.00%)
spills in affected programs: 0 -> 0
helped: 0
HURT: 0

total fills in shared programs: 186 -> 186 (0.00%)
fills in affected programs: 0 -> 0
helped: 0
HURT: 0

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4999>

panfrost: Do fine-grained flushing for occlusion query results

This allows doing occlusion queries in one frame and getting the
results in the next frame without having to flush.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5676>

freedreno/a4xx: fix *_NONE enum conversion

Commit e369b8931c67 ("freedreno: Use explicit *_NONE enum for undefined
formats") only partially converts ~0 to *_NONE enum.  It breaks texture
support, and glmark2 texture scene gives a black screen.

Adding the missing conversion of ~0 to *_NONE enum fixes the issue.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5693>

CI: Re-enable Panfrost T860 jobs

The lab is back online.

This reverts commit 34db50558d9a6dca89218f74c4418cdf0b0acbcb.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5694>

radeonsi: add a debug option to enable NGG culling for tessellation

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5524>

radeonsi: don't try to enable NGG culling for GS

It doesn't do anything.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5524>

radeonsi: always use Wave64 for HS/GS/VS shader stages (except GS fast launch)

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5524>

radeonsi: always use Wave32 for GS fast launch, because Wave64 hangs

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5524>

radeonsi: fix NGG culling for Wave64

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5524>

ac/gpu_info: fix num_physical_sgprs_per_simd for gfx10

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5524>

radeonsi: don't flush in fence_server_sync

This reverts commit 50b06cbc10dbca1dfee89b529ba9b564cc4ea6f6 and fixes
an Android performance regression.

Fixes: 50b06cbc10dbca1dfee89b529ba9b564cc4ea6f6 "radeonsi: fix fence_server_sync() holding up extra work v2"
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5602>

CI: Temporarily disable Panfrost T860 jobs

Phase two of our network reconfiguration is happening this afternoon, so
we need to drop our RK3399 out for a little while. (Part of this
reconfiguration is to shard our devices across networks and racks, so
losing one part of our infrastructure doesn't mean losing any particular
device type.)

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5689>

CI: Re-enable the Windows VS2019 build job

Let's try this and see how it goes.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5689>

CI: Correct build-directory path on Windows, and keep it

Build job artifacts capture Meson logs from _build, so we can analyse
what Meson did during configuration, as well as the full output of any
test jobs.

We were previously calling our build directory 'build', which meant it
wouldn't have been captured by the artifacts, and we were also deleting
it to make really sure there was no chance of logs getting captured
either.

Rename the build directory to '_build' to match the others, and don't
delete it either, so we can keep our configure/test logs.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5689>

CI: Try shared libraries on Windows

This might make linking a bit less prone to OOM when trying to pull in
LLVM.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5689>

CI: Enable assertions on Windows

Getting assertion failures is helpful to have, even if we are doing a
release build.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5689>

radeonsi: bump SI_NUM_SHADER_BUFFERS to 32

Some app uses more than 8 SSBOs (https://gitlab.freedesktop.org/mesa/mesa/-/issues/2946),
so increase SI_NUM_SHADER_BUFFERS to 32 (which allows 16 SSBOs).

Since we're now using a 64 bits number to track buffers, we could bump
SI_NUM_SHADER_BUFFERS to 48 but that would conflict with Mesa's
MAX_COMBINED_ATOMIC_BUFFERS limit (= 90).

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2122
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5632>

glsl: remove stale FIXME

This is no longer an issue, was likely fixed years ago.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5657>

st/glsl_to_nir: disable st_nir_lower_builtin() when packing supported

There is no need to lower builtins when uniform packing is
supported by the driver. Lowering is only required by other drivers
because we prepack builtin uniforms.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3140
CC: <stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5656>

glsl: define gl_LightSource members in ARB_vertex_program order

GLSL shares functionality with ARB_vertex_program but the GLSL
spec defines the gl_LightSource builtin with a member order that
is different from the packing expected in ARB_vertex_program.
This difference introduces a need for specialist lowering code
when handling builtin structs that is not required for normal
uniform structs due to member location mismatches.

Since gl_LightSource can't be redefined it shouldn't matter if
we add the members in the order listed in the spec, just so long
as we add them all. So here we rearrange the definition of the
glsl builtin to reflex our internal layout and that of
ARB_vertex_program. This required for the following patch.

CC: <stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5656>

mesa: add _mesa_program_state_value_size() helper

This allows us to query the uniform size required to store the
state value.

CC: <stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5656>

mesa: remove _mesa prefix from static function

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5656>

zink: set lower_uadd_carry in nir options

fixes a bunch of mulextended piglit tests

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5685>

loader/dri3: Check for window destruction in dri3_wait_for_event_locked

If the underlying X11 window gets destroyed, the event we're waiting
for may never be delivered, in which case xcb_wait_for_special_event
would hang indefinitely.

Solution:

1. Use xcb_poll_for_special_event to check if an event has arrived yet.
2. If not, Wait up to ~1s for XCB's file descriptor to become readable;
   if it does, go back to step 1.
3. If the file descriptor didn't become readable, make a round-trip to
   the X server to check that the window still exists. Go back to step
   1 if it does, otherwise bail.

Also add an early bail-out when it's known that the window was
destroyed.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/116
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5368>

loader/dri3: Use dri3_wait_for_event_locked in loader_dri3_wait_for_msc

Before, if one thread ended up waiting in dri3_wait_for_event_locked
and another one in loader_dri3_wait_for_msc at the same time, one thread
could end up processing an event the other thread was waiting for, which
could result in the latter thread waiting longer than necessary
(possibly indefinitely).

Noticed by inspection.

v2:
* Drop xcb_flush call from loader_dri3_wait_for_msc in favour of the one
  in dri3_wait_for_event_locked (Kenneth Graunke)

Fixes: 7b0e8264dd21 "loader/dri3: Try to make sure we only process our
                     own NotifyMSC events"

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5368>

loader/dri3: Add dri3_wait_for_event_locked full_sequence out parameter

Preparation for the next commit, no functional change intended.

Cc: mesa-stable
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5368>

v3d: Fix -Wmaybe-uninitialized compiler warning in the v33 code.

We weren't initializing the VCM bits in the !gs path, but v33 doesn't have
GS so we can just mark it unreachable.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2952>

v3d: Enable PIPE_CAP_TGSI_TEXCOORD.

Dave wants to drop the !TEXCOORD path from NIR, and it's easy enough to
do.  Untested.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2952>

vc4: Enable PIPE_CAP_TGSI_TEXCOORD.

Dave wants to drop the !TEXCOORD path from NIR, and it's easy enough to
do.  Untested.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2952>

gallium/util: Add a helper function for point sprite handling.

Many drivers will need to do the same thing here, so consolidate it.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2952>

turnip: enable depthBiasClamp

Passes at least dEQP-VK.dynamic_state.rs_state.depth_bias_clamp

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5678>

turnip: enable largePoints

Passes dEQP-VK.rasterization.primitive_size.points.point_size_*

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5678>

freedreno/regs: add extra bits for UBWC array pitch

This is not completely tested, but matches the max array pitch allowed by
A6XX_TEX_CONST_9_FLAG_BUFFER_ARRAY_PITCH.

Note this still doesn't allow all image sizes, but it allows 16384x16384
cpp=4 images to work.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5678>

frontends/va: Handle dynamic resolution/SVC for VP9

VP9 allows frame to use another resolution frame as reference
frames so updating the resolution for decoder when there is a
resolution change.

Signed-off-by: Satyajit Sahu <satyajit.sahu@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5646>

v3d/compiler: fix spill offset

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Fixes: 97566efe5cac0ff11b ("v3d: Rematerialize MOVs of uniforms instead of spilling them.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5664>

radv: add support for MRTs compaction to avoid holes

SPI_SHADER_COL_FORMAT allocates export memory and CB_SHADER_MASK
map them to higher MRTs if necessary. The hardware allows to remap
MRTs to avoid holes somehow.

For example, if we have a scenario where MRT0 is unused and only
MRT1 and MRT2 are used, SPI_SHADER_COL_FORMAT is 0x77 and
CB_SHADER_MASK/CB_TARGET_MASK are 0x770 (this assumes
SPI_SHADER_UINT16_ABGR is set).

This allows us to remove one workaround that was added for fixing
GPU hangs with DXVK. I think this is because SPI_SHADER_COL_FORMAT
expects contiguous MRTs to be allocated.

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/5434>

radv: use SPI_SHADER_ZERO for non-written color attachments

When colorWriteMask is 0 we can assume that this color attachment
is unused.

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/5434>

radv: rework 8/16-bit color attachment formats detection

To prepare for MRTs compaction.

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/5434>

radv: adjust CB_SHADER_MASK for dual-source blending in the shader info pass

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/5434>

radv: enable VK_AMD_shader_ballot on GFX6-7 with both compiler backends

It gives +1-2 FPS with Doom Eternal on Pitcairn.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5659>

nir: Add new rules to optimize NOOP pack/unpack pairs

nir_load_store_vectorize_test.ssbo_load_adjacent_32_32_64_64 expectations
need to be fixed accordingly.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5589>

gallivm/nir: fix const loading on big endian systems

The code was expecting the lower 32-bits of the 64-bit to be
what it wanted, don't be implicit, pull the value from the union.

This should fix rendering on big endian systems since NIR was
introduced.

Fixes: 44a6b0107b37 ("gallivm: add nir->llvm translation (v2)")
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5677>

freedreno/ir3: fix resinfo wrmask

resinfo always writes 3 components, which was not being taken into account

Fixes these tests:
dEQP-VK.renderpass.suballocation.attachment_sparse_filling.input_attachment_3
dEQP-VK.renderpass.suballocation.attachment_sparse_filling.input_attachment_7

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5674>

docs: use ref-links for internal references

Ref-link have two benefits over generic links:

1. They produce the right result for non-HTML outputs
2. They get validated at build-time

So let's use them for internal references instead.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5671>

docs: fix internal references

It seems last time I tried to fix these, I missed a few spots. So let's
try to get things right this time.

Fixes: 429ff054917 ("docs/relnotes: update internal references")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5671>

docs: restore accidentally dropped labels

These were accidentally dropped when cleaning up the TOC, making links to
them dead. Because we used plain links, sphinx didn't inform us that
these became dead. Let's restore them.

Fixes: 14f2a81b6f6 ("docs: drop open-coded toc for articles")
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5671>

docs: remove non-existent reference

The section referenced here was removed a while ago, but it was always
empty anyway. Let's just remove it instead of trying to fix it up.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5671>

meson: Do not require shader cache for radv.

We fixed the compile error a while ago.

Fixes: cc10b34e9ed "util/disk_cache: Fix disk_cache_get_function_timestamp with disabled cache."
Reviewed-by: Drew Davenport <ddavenport@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5649>

rbug: Fix rbug_delete_vs_state lock acquisition.

Fix warning reported by Coverity Scan.

Double unlock (LOCK)
double_unlock: mtx_unlock unlocks rb_pipe->call_mutex while it is
unlocked.

Fixes: 07838ff990a7 ("rbug: Use the call mutex")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3023
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Jakob Bornecrantz <jakob@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5196>

v3d: moving v3d simulator to src/broadcom

So it could be used by both the OpenGL and the Vulkan driver.

In addition to the move, some small changes were needed to be made on
the API. For example, the simulator was receiving v3d_screen on
initialization, and that code setted v3d_screen->sim_file. Now it
returns the new sim_file created.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5666>

turnip: Put VK_KHR_external_fence_fd stubs back

tu_ImportFenceFdKHR is used by tu_AcquireImageANDROID, which may or
may not work, but let's at least keep things compiling until somebody
has time to tie up the loose ends on the Android side.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5670>

iris: Implement pipe->texture_subdata directly

Chris Wilson noted that u_default_texture_subdata's transfer path
sometimes results in wasteful double copies.  This patch is based
on an earlier path he wrote, but updated now that we have staging
blits for busy or compressed textures.

Consider the case of idle, non-CCS-compressed, tiled images:

The transfer-based CPU path has to return a "linear" mapping, so upon
map, it mallocs a temporary buffer.  u_default_texture_subdata then
copies the client memory to this malloc'd buffer, and transfer unmap
performs a tiled_memcpy to copy it back into the texture.  By writing
a direct texture_subdata() implementation, we're able to directly do
a tiled_memcpy from the client memory into the destination texture,
resulting in only one copy.

For linear buffers, there is no advantage to doing things directly, so
we simply fall back to u_default_texture_subdata()'s transfer path to
avoid replicating those cases.

We still may want to use GPU staging buffers for busy destinations
(to avoid stalls) or CCS-compressed images (to compress the data),
at which point we also fall back to the existing path.  We thought
to try and use a tiled temporary, but this didn't appear to help.

Improves performance in x11perf -shmput500 by 1.96x on my Icelake.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2500
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3818>

turnip: Properly return VK_DEVICE_LOST on queuesubmit failures.

The device lost support closely matches the anv code for the same.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2769>

turnip: Fix error handling of DRM_MSM_GEM_INFO ioctls.

drmCommandWriteRead gives us a -errno, and we only checked for -1 (-EPERM,
incidentally).  All the callers wanted 0 for errors, which they were
getting by the fact that req.value was 0-initialized in our stack
allocation (though this only works as long as the kernel doesn't return an
error after setting req.value to something), and -EPERM not really being
an answer we would expect from an ioctl at this stage in the driver.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2769>

turnip: Do better TU_DEBUG=startup logging of drmGetDevices2() failure.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2769>

turnip: semaphore support.

There is only one queue for now, so for non-shared semaphores, the
implementation is basically a no-op. For shared semaphores, this
always uses syncobjs. This depends on syncobj support in the msm
kernel driver.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2769>

ci/baremetal: Bump the kernel to a recent drm-msm-fixes for msm semaphores.

We need this to test the new VK feature we're about to land.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2769>

aco: fix partial copies on GFX6/7

While we don't allow partial subdword copies,
we still need to be able to split 64bit registers

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5663>

mapi: x86: Fix dynamic entries in x86 tsd stubs.

We need to update dynamic entries related code after updating
asm stubs.

Fixes: 45206d7673a ("mapi: Adapted libglvnd x86 tsd changes")
Signed-off-by: Lepton Wu <lepton@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5598>

ci/bare-metal: Fail early when we get stuck powering on a cheza.

I think I've seen about 3 of this error total so far, but waiting 60
minutes for the scripts to give up wastes marge time.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5651>

freedreno/ir3: move nir finalization to after cache miss

In cases where every variant is a shader-cache-hit, we never need the
post-finalize round of nir opt/lowering passes.  So defer this until
the first shader-cache-miss to avoid doing pointless work.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

freedreno/ir3: disk-cache support

Adds a shader disk-cache for ir3 shader variants.  Note that builds with
`-Dshader-cache=false` have no-op stubs with `disk_cache_create()` that
returns NULL.

Binning pass variants are serialized together with their draw-pass
counterparts, due to shared const-state.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

freedreno/ir3: build binning variant at same time as draw variant

For shader-cache, we are going to want to serialize them together.
Which is awkward if the two related variants are not compiled together.

This also decouples allocation and compile, which will simplify adding
shader-cache (which still needs to allocate, but can skip compile).

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

freedreno/a6xx+ir3: stop generating pointless binning shaders

Currently we always do sysmem if there is tess.  And for GS, the binning
pass VS ends up identical to the draw pass VS, so no point in compiling
it twice.  (For GS what we should do someday is generate a binning pass
GS, and possibly if we can do cross-stage linking opts, an optimized
binning pass VS, but the required outputs would somehow have to end up
in the shader variant key.)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

freedreno/ir3: shuffle some variant fields

Just to group together the parts that will get serialized when we have
shader disk-cache.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

freedreno/ir3: add ir3_compiler_destroy()

Use ir3_compiler_destroy() rather than open-coding ralloc_free().  This
will give us a place to add more compiler related cleanup code in the
following patches.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

freedreno/ir3: move finalize_nir to pscreen hook

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

freedreno/ir3: add ir3_finalize_nir()

The next step is to hook this into pscreen->finalize_nir() so it can
come before the state tracker's shader-caching.

Unfortunately we still need to do lower_io after mesa/st, so that is
split out into a post-finalize pass.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5372>

zink: use OpFUnordNotEqual for nir_op_fne

we want to detect NaNs here, and OpFUnordNotEqual is the variant which does this

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

zink: set lower_mul_high and lower_rotate in ntv compiler options

we don't implement these

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

zink: handle isign alu in ntv

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

zink: handle ixor in ntv

fixes spec@glsl-1.30@execution@built-in-functions@fs-op-assign-bitxor tests

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

zink: lower byte/word extract ops in nir

we don't implement these, and pre-optimizing them breaks things in ntv->vtn

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

zink: add bitfield_reverse handling to ntv

fixes several piglit tests

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

zink: add ult handling for ntv

fixes shaders@glsl-vs-absolutedifference-uint piglit test

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

zink: handle signed and unsigned min/max ops in ntv

fixes a number of piglit amd_shader_trinary_minmax tests

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5562>

radv: remove the load/store workaround for Monster Hunter World with LLVM

Now that ACO is default, this is pointless.

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/5658>

radv: remove the shader ballot workaround for Youngblood with LLVM

Now that ACO is default, this is now pointless.

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/5658>

docs: update favicon

I created a new and cleaner favicon for mesa3d.org, and it seems like a
good idea to use that one for the docs as well.

While we're at it, replace the original PNG with the original SVG asset
the ICO-file was generated from.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5643>

turnip: fix huge scissor min/max case

Now that tu_cs_emit_regs is used for the scissor, it hits an assert when
the scissor is too large. Fixes this dEQP test:

dEQP-VK.draw.scissor.static_scissor_max_int32

Fixes: 9c0ae5704d654108fd36b ("turnip: fix empty scissor case")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5655>

turnip: fix VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES

My attempt to be clever here backfired, it overwrites the pNext and stops
the loop (causing deqp to fail to query extension features after that).

Fixes: 62de79ac4492ac9e ("turnip: implement VK_KHR_shader_draw_parameters")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5654>

panfrost: Add PAN_MESA_DEBUG=gl3 flag

This flag allows forcing GL 3.3 without having to use
MESA_GL_VERSION_OVERRIDE etc.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5584>

freedreno/a6xx: use firstIndex field

Analogous to the turnip change.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5644>

tu: Pass firstIndex directly to CP_DRAW_INDX_OFFSET

Saves some minor overhead, cleans things up a bit, and removes one more
unknown. We now program the internal registers in the same way between
direct/indirect draws.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5644>

freedreno/registers: Label firstIndex field in CP_DRAW_INDX_OFFSET

Based on comparing the implementations of CP_DRAW_INDX_OFFSET and
CP_DRAW_INDIRECT, this is what this field is for.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5644>

freedreno: On a5xx+ INDX_SIZE is MAX_INDICES

This was already done correctly for the indirect variants, and turnip
was setting the correct value, but it seems freedreno missed the change
in the non-indirect variant. Also, fix a misspelling of "indices" and
add a type to INDX_SIZE.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5644>

freedreno: Share constlen between different stages properly

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5607>

freedreno: Refactor ir3_cache shader compilation

Use an array, which makes it more like turnip and makes implementing the
const limits easier.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5607>

tu: Share constlen between different stages properly

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5607>

ir3: Add ir3_trim_constlen()

This provides the policy for how to handle reducing constlen for some
stages.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5607>

ir3: Support variants with different constlen's

This provides the mechanism for compiling variants with a reduced
constlen. The next patch provides the policy for choosing which to
reduce.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5607>

ir3: Include ir3_compiler from ir3_shader

I wanted to access the ir3_compiler from a small helper inside
ir3_shader.h, which currently isn't possible.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5607>

ir3, freedreno: Round up constlen earlier

Prevents problems when calculating whether we overflow the shared limit.
Note that on a6xx, the macros handle the assert for us.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5607>

v3d/compiler: don't rewrite unused temporaries to point to NOP register

This was assuming that unused temporaries are written but never read,
since the NOP register can only be used as a destination register,
but we can end up here also for temporaries that are read once but
never written.

This was found with a graphicsfuzz test that has a switch with
cases that have unreachable discards. In that test, NIR genrates
code like this:

decl_reg vec3 32 r19
...
r20 = mov r19.z
r21 = mov r19.y
r22 = mov r19.x

Where r19.xyz would generate 3 temporary registers that are read but
never written, so we would rewrite them to point to the NOP register
as QPU instruction sources, which is not allowed and would hit an
assert that expect magic reads to be from [r0,r5] only.

Fixes:
dEQP-VK.graphicsfuzz.unreachable-switch-case-with-discards

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5645>