radv: regroup dynamic states initialization

It should be possible to initialize these inside libraries at some
point.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20960>

radv: ignore all CB dynamic states when there is no color attachments

This simplifies radv_init_dynamic_state() slightly.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20960>

ci: Run our manual jobs during the nightly scheduled run.

We have a common pain point with fractional CTS coverage, where the test
list changes on a CTS uprev or board load rebalancing, so you get a
different subset of tests run.  The dev updates the list of xfails (a
pain), but also we end up with xfails left behind that aren't tested any
more and don't reflect reality.

For some drivers (tu, freedreno, zink-anv) we have manual jobs available
for curious devs to look at the current state of the CTS, but without
anyone having to keep the full xfails updated during uprevs, you don't
necessarily know what to do with the results you get on your MR.

So, let's introduce nightly testing for the tests that aren't guaranteed
green by Marge.  With that, Someone (possibly me? sigh) can review the
nightly results and push up updates for full-run xfails so everyone can be
on the same page other than a day or so of delay.  We also have some hope
for automated tooling to do this thanks to what Collabora has been working
on for automated CI uprev MR generation.

Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20950>

ci/zink: Move the zink-anv-tgl manual full run to custom manual deps.

Follow-up to !17445, since this run had been added while that MR was
outstanding.  Now it shouldn't show up in unrelated pipelines.

Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20950>

ci/zink: Disable Amnesia trace until the linked issue gets fixed.

Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20950>

radv: Shift left the tile swizzle more on GFX11.

ac/surface puts the raw pip_bank_xor there, which needs the extra
shift for the actual tile_swizzle.

(I think long term we should refactor this in ac/surface but for
 now lets fix like radeonsi to avoid race conditions.)

CC: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20979>

radv: Set FDCC_CONTROL SAMPLE_MASK_TRACKER_WATERMARK

Might cause hangs according to comments. Syncs with radeonsi/PAL.

Fixes: e210ffb4d0d ("radv: update framebuffer registers on GFX11")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20979>

mailmap: Remap e-mail addresses for Faith Ekstrand

Acked-by: Lina Versace <lina@kiwitree.net>
Acked-by: Daniel Stone <daniels@collabora.com>
Lovingly-reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21000>

broadcom/compiler: fix indentation at v3d_nir_lower_image_load_store

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20972>

dzn, driconf: Add a driconf entry for NMS to claim wide line support

Reviewed-by: Giancarlo Devich <gdevich@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20959>

iris: Stop marking context unconditionally as guilty

With this change, we would fetch the reset status and if the context
status is banned or in unknown state, we would just start with the fresh
context.

Also, use the fetched reset status to communicate back to the gallium
frontend.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7802

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/20928>

CI: Lima farm is offline

Acked-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20995>

docs: Add GL 4.3 support info in mesa docs

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20964>

radeonsi/vcn: validate fence handle before using it

Check if picture fence handle pointer is valid before using.

Fixes: 843bdd22 ('radeonsi/vcn: check fence before destroying dpb')

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20986>

wsi/win32: Use app-provided timeout instead of arbitrary hardcoded value

Prevents returning spurious timeouts when the app wanted to wait
infinitely. Fixes 3DMark Wild Lands which would otherwise attempt
to render/present a buffer it didn't successfully acquire.

Reviewed-by: Giancarlo Devich <gdevich@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20963>

wsi/win32: Always use non-SRGB formats for DXGI

The actual buffer is always created as non-SRGB, and then SRGB views
can be used to render into it. Fixes a crash trying to launch
3DMark Wild Lands

Reviewed-by: Giancarlo Devich <gdevich@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20963>

freedreno/drm: Restart import on zombie race

If we hit the race condition of looking up an already imported BO that
is in the process of being destroyed, the handle will be GEM_CLOSE'd,
meaning that the handle that we just got from the kernel is probably not
valid.  So in this case we should retry.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20961>

freedreno/drm/virtio: Flush before CREATE_BLOB

The RESOURCE_CREATE_BLOB ioctl can carry a ccmd payload, similarly to
EXECBUF.  But we need to preserve the order of buffered execbuf cmds
which haven't been flushed to the guest kernel yet, rather than let the
CREATE_BLOB payload jump to the head of the queue.  Otherwise, for ex,
the host could see the guest requesting an iova that has not yet been
(from it's perspective) released.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20961>

radv: rename radv_create_shaders() to radv_graphics_pipeline_compile()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20943>

radv: split radv_create_shaders() between graphics and compute shaders

This introduces radv_compute_pipeline_compile() which is used for
compute and ray tracing pipelines. I think it's better than having a
single function for compiling everything, and that will allow us to do
more cleanups.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20943>

radv: pass the number of stages to radv_hash_shaders()

This will help for splitting radv_create_shaders().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20943>

radv: simplify VK_PIPELINE_CREATE_FAIL_ON_PIPELINE_COMPILE_REQUIRED

The Vulkan spec says:
    "VK_PIPELINE_CREATE_FAIL_ON_PIPELINE_COMPILE_REQUIRED_BIT specifies
     that pipeline creation will fail if a compile is required for
     creation of a valid VkPipeline object; VK_PIPELINE_COMPILE_REQUIRED
     will be returned by pipeline creation, and the VkPipeline will be
     set to VK_NULL_HANDLE."

Given the implementation is expected to set the pipeline to
VK_NULL_HANDLE, it's unecessary to handle pipeline feedback.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20943>

rusticl: fix build error with valgrind being enabled

This bumps the meson requierement to 1.0 because it requires
https://github.com/mesonbuild/meson/pull/11024

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7688
Fixes: 20c90fed5a0 ("rusticl: added")
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19778>

asahi: Omit extra call to clock_gettime

It's cheap but it isn't free.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20973>

gallium/draw: Enable polygon stipple NIR helpers to generate bool1 or bool32 Booleans

It appears that only softpipe, llvmpipe, and d3d12 hit any of this
code. If some NIR-to-TGSI driver that doesn't have native integers
(e.g., i915 or r300) wants to use this path in the future, it should be
easy to add float32 support.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20869>

gallium/draw: Enable aapoint NIR helpers to generate bool1, bool32, or float32 Booleans

Fixes arb_point_parameters-point-attenuation on G33. The crash in
point-line-no-cull is fixed, but the test still fails.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20869>

ir3: support texture and sampler index with offsets

"If this texture instruction has a nir_tex_src_texture_offset source,
then the texture index is given by texture_index + texture_offset."

This fixes the failures for:
spec@arb_arrays_of_arrays@execution@sampler@fs-nested-struct-arrays-nonconst-nested-array
spec@arb_gl_spirv@execution@uniform@sampler2d-nonconst-nested-array

Signed-off-by: Amber Amber <amber@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20954>

etnaviv: drm: fix check if BO is on a deferred destroy list

list_is_linked() isn't the right function to use in order to check if
the BO is on a cache bucket or the zombie list, as this checks if the
next pointer of the list isn't NULL. This is always the case with the
BO list item as it's always initialized, so the next pointer points to
the list head itself when the BO isn't on any list.

Use list_is_empty() to check if the BO is actually linked into one
of the deferred destroy lists.

Fixes: 1b1f8592c03c ("etnaviv: drm: properly handle reviving BOs via a lookup")
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20940>

etnaviv: free pm queries dynarray on screen destroy

CC: mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20940>

etnaviv: drm: fix BO array leaks

Free the both arrays tracking BOs when the etna_cmd_stream is destroyed.

CC: mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20940>

anv: fix null descriptors

When writing descriptor with a null buffer/image we expect that
writing 0 will point to the null surface. For that to work the null
surface has to be in the bindless surface heap.

This fixes some new failures in dEQP-VK.robustness.* tests once
rewritten from the NV_ray_tracing to KHR_ray_tracing extension.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 4ceaed7839 ("anv: split internal surface states from descriptors")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7762
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20953>

turnip: make debug_flags a global variable

Add tu_env as a global variable and add tu_env_init to initialize it.
Add TU_DEBUG macro to check debug flags.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20946>

turnip: replace TU_DEBUG_DONT_CARE_AS_LOAD by a bool

Instead of using TU_DEBUG=dontcare_as_load, it can still be overriden
using vk_dont_care_as_load=true.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20946>

dzn: Consider linked shaders when computing DXIL hash

Fixes 3DMark Wild Lands. Otherwise, we'd end up loading a DXIL shader
that had invalid linkage with another shader in the pipeline. We can
only load a DXIL shader if it's being linked against the same before
and after as a previous compilation.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20962>

pvr: Process wait event sub command.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20862>

pvr: Process set and reset event sub commands.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20862>

zink: set VK_PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT when using DB

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

zink: disable bindless texture ext with descriptor buffer

it's illegal to mix-and-match DB and non-DB descriptor usage, so disable
this until I finish the implementation

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

zink: always set RESOURCE usage for descriptor buffers

all types of descriptors may use buffer-type descriptors, even samplers

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

zink: assert that buffer descriptor usage is populated before bind

this is illegal

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

zink: set vkusage/vkflags for buffer resource objects

this needs to be populated for other places in the driver that need it

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

zink: add VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT for buffers if ext is enabled

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

zink: fix heap/memory type selection

the memory type index still needs to be used for suballocation over the
heap since that's the actual type of allocation being used

Fixes: f6d3a5755f6 ("zink: zink_heap isn't 1-to-1 with memoryTypeIndex")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

Revert "zink: allow direct memory mapping for any COHERENT+CACHED buffer"

This reverts commit a3552048c7e7b8afebfc99caac6d555439ddf2bc.

on some drivers this catches qbos, which then hits the below assert.
needs more investigation

Fixes: a3552048c7e ("zink: allow direct memory mapping for any COHERENT+CACHED buffer")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20912>

anv: there's no need to set exec_obj offsets twice

The anv_execbuf_add_bo() function already sets the offsets for the
exec_objects. Since we're always using softpin and never using
relocations all these objects should have non-changing offsets, all
set during anv_bo creation and never changed. Not only we should not
change these offsets, we definitely don't change them between
anv_execbuf_add_bo() and this loop we're removing.

Previously, we'd have the offset set as -1 for BOs that had never been
submitted when we were not using softpin.

Notice that with games we can have several hundreds of BOs in this
array.

This loop was added by:
  c5f7e1f5b4a1 ("anv: Delete relocation support from batch submission")

Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20885>

radv: Loop over shader stages in flush_indirect_descriptor_sets.

Reduces boilerplate.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20935>

radv: Fix emitting tess indirect descriptors twice.

This fixes a copy-paste error found by manual inspection.

TES may be merged into GS with certain HW stage mappings, which lead to
duplicate set-register commands to be emitted with the old code.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20935>

pvr: Split render job submission for multi-layer framebuffers

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20932>

pvr: Rename global_queue_job_count to global_cmd_buffer_submit_count

This makes the name more accurate, since submits which require multiple
job submissions behind the scenes do not additionally increment this
counter.

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20932>

pvr: Add pvr_csb_bake()

This is a simple helper for minimizing the storage requirements of
control streams. It discards all information required only while
building the control stream and returns just the status and the list of
BOs backing the control stream. The first BO in the list is the start
of the control stream.

Especially for small, deterministically sized control streams, there's
no sense in lugging around an entire builder structure once it's built.

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20932>

aco: set has_color_exports with GPL

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Fixes: 192486b7aa5 ("aco/gfx11: export mrtz in discard early exit for non-color shaders")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20937>

radv: add a layer for fixing rendering issues with RAGE2

This game seems to incorrectly set the render area and since we switched
to full dynamic rendering, the framebuffer dimensions is no longer used.

Forcing the render area to be the framebuffer dimensions restore the
previous logic and it fixes rendering issues.

Fixes: c7d0d328d56 ("radv: Set the window scissor to the render area, not framebuffer")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20900>

radv: Optimize emitting prefetches

Check the need for emitting prefetch before calling si_emit_cache_flush
to mask a possible cache miss delay and always inline radv_emit_prefetch_L2.
Either change alone is not significant but together they increase
drawcall throughput by 8% on i5-2500.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20877>

radv: Avoid redundant fetch of radv_device

0.6% gain in drawcall throughput on i5-2500.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20877>

radv: Clean up variables in si_get_ia_multi_vgt_param

8% gain in drawcall throughput on i5-2500. This can significantly change
how compiler allocates registers.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20877>

radv: Change radeon_cmdbuf counters to uint64_t to make alias analysis optimize radeon_emit better

1% gain in drawcall throughput on i5-2500. Depends quite heavily on
compiler and CPU.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20877>

ci/debian/x86_test-vk: drop an outdated dependency

This was used by the tracing jobs, which we are not running on Mesa
anymore.

Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20956>

trace: Don't use italic escape code.

It's not widely supported, and often emulated with invert highlight,
which is very distracting.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20934>

llvmpipe: Honor zero sample_mask when multisample is disabled.

The JIT generated FS shader has logic to obey sample mask when:
multisample is enabled, or multisample is disabled but FS writes sample
mask and pipe_rasterizer_state::no_ms_sample_mask_out.

However it did not handle the case where multisample was disabled, FS
did not write sample_mask, and sample mask was zero.  Instead it relied
upon the setup to discard the primitives, but that went away with commit
da5840f3.

We could restore the discard on zero mask behavior, but we would again
blurring the semantics of rasterization discard.  Instead this change
adds logic to primitive setup to cull the primitives when sample mask is
zero.

Fixes: da5840f3 ("llvmpipe: Faithfully honour pipe_rasterizer_state::rasterizer_discard flag")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20934>

meson: do not reconstruct ICD paths

Meson will already construct these paths for us, so let's reuse them
instead of throwing away the result and recontstructing them.

Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20907>

freedreno/meson: simplify script-path logic

When passing this to files(), there's no point in prepending
current_source_dir. Let's simplify this a bit.

Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20907>

meson: use files() instead of joining paths

The Meson docs points out that it's better to use the files() function
when referring to files in the source tree than manually constructing
paths like this. Let's follow that advice, and get some neat cleanups.

Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20907>

meson: avoid using deprecated build_root() method

The meson.build_root() method has been deprecated, so let's switch to
meson.project_build_root(), which usually means the same thing. The case
where it doesn't do the same thing is if Mesa is a subproject to some
other project, but in that case I believe we want the build root of Mesa,
not of the parent project anyway.

Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20907>

pvr: small cleanups

- remove unused PVR_IDF_WDF_IN_REGISTER_CONST_COUNT define
- remove unneeded pvr_compute_ctx forward declaration
- add render context section header to match compute and transfer

Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Matt Coster <matt.coster@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20931>

intel/compiler: fine-grained control of dispatch widths

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com> [v2]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20854>

meson: remove deprecated dri-drivers option

This was deprecated in cdde031ac2c ("classic/i965: Remove driver"),
which is almost two years ago, and many major releases ago. Nobody
should be using this any more. Let's remove it.

Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20905>

meson: remove deprecated osmesa-bits option

We've released several new major versions since this was deprecated in
202cab3d308 ("meson: deprecate specifying osmesa-bits"), so it's
probably about time to remove it.

Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20905>

meson: remove dupliace add_devenv call

Looks like a bad merge-conflict resolution.

Fixes: 7bed7d94756 ("meson: use sections in summary()")
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Thomas Andersen <phomes@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20908>

egl/dri2: Use primary device in EGL device platform for kms_swrast

For KMS only devices, there is no render node available. Also remove
unused _eglGetDRMDeviceRenderNode function.

Signed-off-by: Lepton Wu <lepton@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20592>

ci: uprev virglrenderer

This change contains a memory unmap fix and makes the render server mandatory.

Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20938>

egl: Rewrite eglGetMscRateANGLE to avoid probes and handle multi-monitor

RRGetScreenInfo re-probes connector status, which may result in an EDID
transfer for every output, which according to Adam Jackson can be on the
order of 100ms for a single EDID block.  So our previous implementation
of this eglGetMscRateANGLE was blocking for excessive periods of time
instead of being a quick query of the refresh rate like users expect.

This changes our eglGetMscRateANGLE implementation from using
RRGetScreenInfo to RRGetScreenResourcesCurrent and RRGetCrtcInfo.
This obtains the same monitor info without re-probing connectors.

Fixes a severe performance regression in Chromium WebGL performance.

While we're re-implementing the extension, we also implement proper
multi-monitor support: if there are multiple active CRTCs, we determine
which contains the largest portion of the surface, as specified in the
EGL_ANGLE_sync_control_rate extension.

We also now report fractional refresh rates correctly rather than
rounding to the nearest Hz.

Fixes: 47526556494 ("egl/x11: implement ANGLE_sync_control_rate")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6996
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7038
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20665>

loader: Add infrastructure for tracking active CRTC resources

This provides a cached view of the current screen resources, with the
coordinates and refresh rate for every active CRTC.  It's currently only
implemented for X11/XCB.

Fixes: 47526556494 ("egl/x11: implement ANGLE_sync_control_rate")
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20665>

zink: reorder commands more aggressively

by starting resources in the unordered state in a given batch, they
gain more opportunities to be promoted to the barrier cmdbuf and avoid
breaking renderpasses

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20890>

CI/windows: Apply CI_FDO_CONCURRENT to piglit too

Reviewed-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20924>

CI/windows: Don't limit deqp-runner to 4 jobs

If FDO_CI_CONCURRENT is set, use that, otherwise let deqp-runner
choose concurrency based on system CPU cores.

Reviewed-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20924>

ac/llvm: remove llvm:: now that we use "using namespace llvm"

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20297>

ac/llvm: run the IPSCCP pass

AMDVLK runs it and it seems useful.

https://en.wikipedia.org/wiki/Sparse_conditional_constant_propagation

58380 shaders in 35438 tests
Totals:
SGPRS: 2709080 -> 2709224 (0.01 %)
VGPRS: 1592972 -> 1592808 (-0.01 %)
Spilled SGPRs: 2420 -> 2420 (0.00 %)
Spilled VGPRs: 1077 -> 1077 (0.00 %)
Private memory VGPRs: 253 -> 253 (0.00 %)
Scratch size: 1232 -> 1232 (0.00 %) dwords per thread
Code Size: 61382088 -> 61356504 (-0.04 %) bytes
Max Waves: 849293 -> 849308 (0.00 %)
Outputs: 127090 -> 127090 (0.00 %)
Patch Outputs: 579 -> 579 (0.00 %)

Totals from affected shaders:
SGPRS: 5400 -> 5544 (2.67 %)
VGPRS: 6200 -> 6036 (-2.65 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 975824 -> 950240 (-2.62 %) bytes
Max Waves: 1214 -> 1229 (1.24 %)
Outputs: 232 -> 232 (0.00 %)
Patch Outputs: 0 -> 0 (0.00 %)

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20297>

ac/llvm: run the LLVM sinking pass because LLVM will stop running it

shader-db was run with the sinking pass disabled in LLVM.

58380 shaders in 35438 tests
Totals:
SGPRS: 2730768 -> 2730768 (0.00 %)
VGPRS: 1592932 -> 1592928 (-0.00 %)
Spilled SGPRs: 2687 -> 2687 (0.00 %)
Spilled VGPRs: 551 -> 551 (0.00 %)
Private memory VGPRs: 253 -> 253 (0.00 %)
Scratch size: 700 -> 700 (0.00 %) dwords per thread
Code Size: 61238872 -> 61238868 (-0.00 %) bytes
Max Waves: 849209 -> 849209 (0.00 %)
Outputs: 127090 -> 127090 (0.00 %)
Patch Outputs: 579 -> 579 (0.00 %)

Totals from affected shaders:
SGPRS: 440 -> 440 (0.00 %)
VGPRS: 396 -> 392 (-1.01 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 49880 -> 49876 (-0.01 %) bytes
Max Waves: 105 -> 105 (0.00 %)
Outputs: 14 -> 14 (0.00 %)
Patch Outputs: 0 -> 0 (0.00 %)

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20297>

anv: add a third memory type for LLC configuration

Commit 582bf4d9 turned on write-combining for most (all?) memory
allocations.  This caused a fairly large performance drop in some of
our VMware tests (application traces, such as Windows Metro Paint).

This patch adds a third memory type configuration: DEVICE_LOCAL,
HOST_VISIBLE, HOST_COHERENT.  This is uncached.  Then, in
anv_AllocateMemory() we only use write-combining for this uncached
type.  This memory type is found in the Intel Windows Vulkan driver.
And according to
https://asawicki.info/news_1740_vulkan_memory_types_on_pc_and_how_to_use_them
uncached memory correlates to write-combined memory.

This fixes our performance regression (and actually produced the
fastest ever results for our test suite).

Signed-off-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20770>

freedreno/drm: Synchronize handle close and lookup

Handle lookup (for example PRIME_FD_TO_HANDLE) must be synchronized with
GEM_CLOSE, otherwise re-import can race with bo_del path, resulting in
the handle of the newly (re)imported BO getting closed.  Now that the
finalize step has been decoupled, fixing this is mostly just deleting
code.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>

freedreno/drm: Split out bo->finalize()

The complexity around batching up handle closing is simply to allow the
virtgpu to back up ccmd's to the host (because virtio/virtgpu is pretty
inefficient when it comes to lots of small msgs to the host, and it is
common that when we are deleting BOs, we delete a lot of them at the
same time.  But that will make the locking fix in the next commit
impossible (without nested locks).  So let's flip this around and do the
step that virtgpu wants to batch up first, before we get into closing
GEM handles, etc.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>

freedreno/drm: Remove bo_del_or_recycle()

In prep for the next patch, where locking is swapped around to cover the
whole bo_del() path, decouple handling of the recycle-to-BO-cache path.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>

freedreno/drm: Detect zombie BOs

When importing from a GEM name or dmabuf fd, we can race with the final
unref of the same BO, in which case we can get a hit in the handle
table for an fd_bo that another thread is about to free().  Detect and
handle this case.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>

freedreno/drm: Add some ref/unref debugging

Helpful to catch common refcnt issues, like resurrecting a zombie
object.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>

freedreno: Don't sync timestamps while perfetto isn't running.

This may help with the regression in trace perf testing since enabling
perfetto on the test drivers.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20916>

dzn: Report as a software device for non-Windows

Fixes: 5f1b8b3e6c9 ("dzn: Use DXGI swapchains")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20939>

dzn: Don't recursively lock the physical device enum mutex

Fixes: cfa260cd274 ("dzn: Use common physical device list/enumeration helpers")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20939>

dzn: Fix Windows WSI

This was a merge conflict from the Win32 WSI DXGI swapchain changes.
I missed moving a new line of code that was added when rearranging
things for using the common helpers.

Fixes: cfa260cd ("dzn: Use common physical device list/enumeration helpers")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20944>

docs: update calendar for 22.3.4

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20942>

docs/relnotes: add sha256sum for 22.3.4

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20942>

docs: add release notes for 22.3.4

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20942>

freedreno/registers: Add RBBM_GPR0_CNTL for non-GMU operation

Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20925>

freedreno/a6xx: Add a few kernel regs/etc

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20925>

virgl: remove unused virgl_encoder_inline_write

The only user was removed with

be8eeb3b59abf18d6b706016c5fa05b746d88f65
  virgl: remove unused virgl_transfer_inline_write

so drop this code too.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18543>

ir3, isaspec: add raw instruction to assembler/disassembler.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20789>

frontends/va: revert commit 0b02db30

revert commit 0b02db30 as it is not a
proper way to fix av1 decoding 10bit issue.

this is corresponding to the fix in
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20870

Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20911>

kopper: Do not free the given screen in initScreen implementation

The given screen is already freed by the caller in case a NULL-pointer is
returned by the implementation.

Cc: mesa-stable
Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20933>

util/tests/cache_test: Skip Cache.List if not supported

FOZ_DB_UTIL_DYNAMIC_LIST depends on inotify support

Fixes: 3b69b67545b ("util/fossilize_db: add runtime RO foz db loading via FOZ_DBS_DYNAMIC_LIST")

Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20865>

util/fossilize_db: add ifdef for inotify header

FOZ_DB_UTIL_DYNAMIC_LIST is defined if the inotify header was
detected.

Fixes: 3b69b67545b ("util/fossilize_db: add runtime RO foz db loading via FOZ_DBS_DYNAMIC_LIST")

Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20865>

nir: Add algebraic optimization for VKD3D-Proton fp32->fp16 conversion.

VKD3D-Proton DXBC f32 to f16 conversion implements a float conversion using PackHalf2x16.
Because the spec does not specify a rounding mode, it emits a sequence to ensure
D3D-like behaviour for infinity.

When we know the current backend has pack_half_2x16_rtz_split,
we can eliminate the extra sequence.

Fossil DB stats on GFX11:
Totals from 835 (0.62% of 134913) affected shaders:
VGPRs: 49368 -> 49224 (-0.29%)
CodeSize: 5341956 -> 5124564 (-4.07%)
Instrs: 1024062 -> 987041 (-3.62%)
Latency: 6530956 -> 6465120 (-1.01%); split: -1.01%, +0.00%
InvThroughput: 908189 -> 870253 (-4.18%)
VClause: 18704 -> 18702 (-0.01%); split: -0.02%, +0.01%
SClause: 33406 -> 33284 (-0.37%); split: -0.38%, +0.01%
Copies: 67440 -> 65992 (-2.15%); split: -2.15%, +0.00%
Branches: 18498 -> 18465 (-0.18%)
PreSGPRs: 38409 -> 38331 (-0.20%)
PreVGPRs: 44089 -> 43834 (-0.58%)

Note, some fossils are from before this pattern was added to VKD3D-Proton,
so the above may not reflect real-world impact.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15838>