agx: Don't crash trying to encoding minifloats

Fixes assertion fails in piglit isinf-and-isnan, which uses a constant infinity,
which has an out-of-bounds mantissa (but the function contract says that's
fine and we just return something undefined.)

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20563>

agx: Use BITFIELD64_BIT for outputs_written

Fix by inspection.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20563>

nir/gather_info: Handle store_zs_agx

This acts as a depth/stencil write. The AGX compiler checks outputs_written to
determine what conservative depth settings the driver needs. Nominally, this
should work: the original store_output(FRAG_RESULT_DEPTH) intrinsic causes the
DEPTH outputs_written bit to be set, so the metadata is still correct after
lowering store_output to store_zs_agx. However, there are a handful of places
that call nir_gather_info late, which *resets* the existing outputs_written
value and regathers, causing Asahi to use the wrong conservative depth settings
when shuffling NIR pass order and breaking gl_FragDepth.

To fix, handle store_zs_agx conservatively when gathering info so we don't have
to play games with the pass order or stashing info in a sideband.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20563>

agx: Wire up nir_intrinsic_store_agx

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20558>

agx: Remove load_global(_constant) support

Now lowered in NIR to better instructions than we were selecting.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20558>

agx: Use load_global_constant for UBO lowering

Rely on the common address arithmetic optimizations. We don't need the
special formats for UBO loads anyway, so this is simpler and optimizes
out the ushr.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20558>

agx: Lower global loads/stores to AGX versions

This lets us do all the needed address arithmetic in a central place.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20558>

nir: Add store_agx intrinsic

This works like store_global, but lets us optimize address arithmetic. Like
load_agx, it is formatted to match the hardware semantic. We don't make use of
any clever formats in this series, though.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20558>

freedreno/a6xx: Demote coherent img access to uncompressed

We cannot use UBWC for coherent access across shader stages.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20612>

gallium: Add image volatile/coherent flags

Freedreno needs to know when an image has volatile or coherent access
flags in the shader.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20612>

mesa/st: Track complete access qualifier for images

Don't turn gl_access_qualifier coming from NIR back into GL enums,
losing information in the process.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20612>

d3d12: implement the get_decoder_fence vfunc

Implement the get_decoder_fence vfunc. Note that the waiting for
completion in this driver happens in the end_frame vfunc itself.

Signed-off-by: Sil Vilerino <sivileri@microsoft.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

radeonsi: vcn: implement the get_decoder_fence vfunc

Implement the get_decoder_fence vfunc by waiting on the fence
previously passed in the end_frame vfunc.

Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

radeonsi: uvd: implement the get_decoder_fence vfunc

Implement the get_decoder_fence vfunc by waiting on the fence
previously passed in the end_frame vfunc.

Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

r600: uvd: implement the get_decoder_fence vfunc

Implement the get_decoder_fence vfunc by waiting on the fence
previously passed in picture->fence in the end_frame vfunc.

Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

virgl: video: implement the get_decoder_fence vfunc

Implement the get_decoder_fence vfunc by waiting on the fence
previously passed in picture->fence in the end_frame vfunc.

Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

frontends: va: fix vaQuerySurfaceStatus and vaSyncSurface for decode

Use the new get_decoder_fence vfunc to  implement
vaQuerySurfaceStatus and vaSyncSurface in the va state tracker.

A pointer to the surface's fence is passed to the codecs before the
end_frame vfunc and the codec is responsible for allocating a fence on
command stream submission.

This fence is then queried on vaQuerySurfaceStatus and waited on in
vaSyncSurface.

Notably both functions were not implemented as per the VA-API docs for
PIPE_VIDEO_ENTRYPOINT_BITSTREAM.

Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

gallium: pipe: add PIPE_DEFAULT_DECODER_FEEDBACK_TIMEOUT_NS

Add PIPE_DEFAULT_DECODER_FEEDBACK_TIMEOUT_NS as a way to control
how much to wait for decoders if this is supported.

Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

gallium: pipe: add get_decoder_fence vfunc

Add a get_decoder_fence vfunc that can be used to query the status
of the previous decode job denoted by 'fence' given 'timeout'.

A pointer to a fence pointer can be  passed to the codecs before the
end_frame vfunc and the codec should then be responsible for allocating
a fence on command stream submission.

Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>

freedreno/a6xx: Remove shader key from fd6_emit

Only construct the key on-demand if the PROG state is dirty.  The newly
added "virtual" PROG_KEY state is used to know when other state that the
shader key depends on changes.  Worth ~13% at drawoverhead test 0.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>

freedreno/a6xx: Avoid zero-init fd6_emit

A pretty significant amount of time spent in fd6_draw_vbo is calling
memset to zero init the on-stack struct.  And a big part of the size
of the struct is fd6_state, of which we only need to initialize
num_groups to zero.  This is worth a 15% improvement in drawoverhead
test 0 ("no state change").

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>

freedreno/a6xx: Tighten up fd6_emit

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>

freedreno/a6xx: Move fb_read descriptor to fd6_gmem

We already overwrote the entire descriptor in patch_fb_read_sysmem().
Doing the same in patch_fb_read_gmem() will simplify things for moving
the fb_read descriptor to the FS's bindless descriptor set.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>

freedreno/a6xx: Split out fd6_state

Split out the build-up of CP_SET_DRAW_STATE packet, as we are going to
want to re-use this for compute state later when we switch to bindless
IBO descriptors.

While we are at it, drop the enable_mask param, as this is determined
solely by the group_id, and it is easier to maintain a table for the
handful of exceptions to ENABLE_ALL.  The compiler should be able to
optimize away the table lookup.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>

freedreno/a6xx: Rename fd6_emit_3d_state()

To better reflect this is emitting 3d state, and to avoid a name
conflict in the next patch.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>

wsi/meson: add missing dep

Same thing as https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20530:
newly added `src/vulkan/util/rmv/vk_rmv_tokens.h` (see !17331) includes
`src/util/` files, so anything that includes it needs `idep_mesautil`.

    In file included from ../src/vulkan/util/rmv/vk_rmv_common.h:29,
                     from ../src/vulkan/runtime/vk_device.h:26,
                     from ../src/vulkan/wsi/wsi_common.c:31:
    ../src/util/simple_mtx.h:34:12: fatal error: valgrind.h: No such file or directory
       34 | #  include <valgrind.h>
          |            ^~~~~~~~~~~~
    compilation terminated.

Fixes: 5f30a7538befb00e7989 ("vulkan: Add RMV token definitions")
Signed-off-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20642>

radv,aco: fix enable_mrt_output_nan_fixup for RAGE2 again

Driver workarounds for game bugs can be easily broken. This one
shouldn't be applied to meta shaders and this restores previous logic.

Fixes: da32cbb5c69 ("aco: fix missing uses of MRT output flags")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20637>

ci: debian-build-testing, build only the LTO variant

The chance we'll miss anything from non-LTO is minimal, and having
both builds in one is too slow (usually the latest job to finish).

Acked-by: Martin Roukala <martin.roukala@mupuf.org>
Acked-by: Eric Engestrom <eric@igalia.com>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20623>

docs: fixup envvar-directive syntax

This directive needs a newline following it to render correctly.

While we're at it, fixup the incorrect indent for one of the
descriptions.

Fixes: 0c58ad3e321 ("docs: use envvar directive more")
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20640>

docs: use envvar directive more

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20638>

v3dv: expose VK_KHR_shader_integer_dot_product

NIR will automatically lower all of these opcodes unless the driver
specifies that it can handle them natively. We don't have any hardware
support for any of these opcodes though, so we just let NIR lower
all of them.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20639>

v3d/compiler: remove unused sample_coverage field from fs key.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20634>

v3dv: fix alpha-to-one for single sample setup

In Vulkan this is expected to work with single sample scenarios too.

Fixes new test in CTS main:
dEQP-VK.pipeline.monolithic.multisample.alpha_to_one.samples_1

cc: mesa-stable

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20634>

v3d: add a debug option to optimize shader compile times

Particularly, this makes compilation stop as soon as we get a
valid shader and doesn't try to optimize spilling by trying
fallback strategies.

Might come in handy to reduce CTS execution time, for example,
dEQP-VK.ssbo.layout.random.8bit.all_per_block_buffers.6 goes from
43m46.715s down to 15m15.068s.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20601>

dzn: Allow multiple graphics queues

There's no real reason not to, WDDM supports it. It's not really that
useful, but I don't expect most apps to really want to do it anyway.
It does enable some useful synchronization scenarios sometimes.

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

dzn: Use DXGI swapchains

Makes things so much faster than doing CPU copies using StretchBlt().

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Hook-up DXGI swapchains and DComp

The CPU copy is horribly slow, so let's hook-up DXGI swapchains. Note
that we're still limited in term of features. For instance, we can't
support more than 2 images per swapchain because of the DXGI present
ordering constraint. We also have to do an extra copy, because DXGI
only allows rendering to a resource on the queue that the swapchain
was created against, but swapchains in Vulkan don't have a queue.

The swapchain is bound to the window using DirectComposition aka
DComp. The DComp infrastructure is set up in the surface, and is
transitioned from one swapchain to the next when the new swapchain
begins presenting.

Unlike Wayland and X, there's no requirement that the compositor has
to release a surface before you can start rendering against it. However,
since we're now supporting the non-sw path, we do need to prevent apps
from rendering to a resource *while* the blit is occurring. We do this
by blocking for a fence while acquiring an image.

Co-authored-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Port Win32 WSI to C++

In a future commit we'll want to be able to use DirectComposition
in the Win32 WSI layer, and unfortunately DComp doesn't have a C-
compatible header (despite being COM, it uses function overloads
and has references instead of pointers for some functions).

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Allow platform code to define extra dependencies

Needed for the windows platform if we want to hook up DXGI swapchains.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Add support for image -> image blits

The win32 swapchain can be backed by a DXGI swapchain, but such swapchains
are incompatible with STORAGE images (AKA UNORDERED_ACCESS usage in
DXGI). So, we need to allocate an intermediate image that will serve as
a render-target, and copy this image to the WSI image when QueuePresent()
is called. That's pretty similar to what we do for the buffer blit case,
except the image -> buffer copy is replaced by an image -> image copy.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Dissociate the blit context and image configuration steps

Even if all implementations wsi_configure_buffer_image() will need to
configure the image as well, it feels a bit weird to call
wsi_configure_image() from there, so let's the
wsi_configure_buffer_image() users call wsi_configure_image() too.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Prepare things for image to image blits

Right now, the WSI core supports copying WSI images to a linear buffer
for implementations that want the result in this form. This being said,
most of the blit logic can be re-used for image to image copies, and that's
exactly what we'll need if we want to hook-up DXGI swapchains in the
win32 WSI implementation. So let's rename a few fields so we no longer
imply that images are copied to a buffer, and the use_buffer_blit boolean
an enum so we can extend the implementation to support image -> image
copies.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Don't set use_buffer_blit manually in the win32 WSI logic

We can simply pass use_buffer_blit=true to wsi_swapchain_init().

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Simplify wsi_win32_surface_create_swapchain() error path

We are about to add more to this function, so let's try to automate
the cleanup steps in the error path. Incrementing image_count as
we add new images to the swapchain allows us to call
wsi_win32_swapchain_destroy() even if not all images were initialized.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

vulkan/wsi: Indent fixes in ws_common_win32.c

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

meson: Define VK_USE_PLATFORM_WIN32_KHR for C++

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>

docs: update new_features.txt for additional work with RADV

This release (23.0) contains less features than usual but the last
branchpoint was delayed and we had end of year holidays.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20609>

intel/compiler: fine-grained control of dispatch widths

Reviewed-by: Matt Turner <mattst88@gmail.com> [v1]
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20535>

ci/venus: re-enable venus-lavapipe

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19542>

ci/venus: specify render server path to enable server config

The render server path is required by crosvm to enable render server.

v2: updated script to package server binary properly

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Corentin Noël <corentin.noel@collabora.com> (v1)
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19542>

ci: uprev virglrenderer

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Corentin Noël <corentin.noel@collabora.com>
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19542>

Fix incorrect stage reference

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20616>

d3d12: Unionize shader variables in d3d12_shader_key to reduce size

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20616>

docs: Add short documentation about RMV tracing variables

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

radv: Enable RMV tracing

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

radv: Add RMV resource tracking

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

radv: Add RMV tracing layer

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

radv: Add RMV tracing utilities

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

radv: Use internal allocation helpers for internal allocations

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

vulkan: Add RMV file exporter

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

vulkan: Add common RMV tracing infrastructure

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

vulkan: Add RMV token definitions

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17331>

zink: set surface->dt when updating swapchain

this otherwise re-creates swapchain surfaces on every frame and has
a significant perf hit for no reason

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20613>

d3d12: Reuse sampler tables inside of a batch

If a shader's sampler state is dirty often, the sampler descriptor heap
can get used up quickly, forcing flushing. If that happens quickly, we
run out of batches and have to wait for batches to finish on the GPU.

When this happens, it is often because the sampler state is switching,
not because it's truly unique. This change hashes and saves sampler
descriptor tables that can be reused in subsequent draws in the same
batch, instead of re-copying the same descriptors and consuming the
heap.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20618>

svga: fix max number of texture cube levels

The maximum number of mipmap levels supported for cubemap can be
determined from the maximum 2D texture size. There is no need
to limit the max to 12.

This fixes a regression in creating GL4.1 and up context since
commit 2658d025166c is now explicitly checking for
MaxCubeTextureLevels >= 15 for GL4.1 context.

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20600>

anv: Print the BO sizes in KB instead of hex bytes.

We already show the address range, which is most of why I'd think you'd be
looking at hex values.  I find a more human-readable number nice for
debugging, instead of counting zeroes to decide if it's 1.5MB or 96kb.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20540>

anv: Fix the size/aperture space debug printouts to consider _ccs_size.

It's added in at anv_bo_vma_alloc_or_close(), so count it here too.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20540>

anv: Add an aperture space summary to INTEL_DEBUG=submit.

Same as on iris, this is nice for tracking at a high level how much memory
is being used.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20540>

r300: allow more copy propagation with relative addressing

We are now extremelly careful when copy propagating a mov that uses
relative addressing. The search for readers will trigger abort when it
sees any other instruction using a relative addressing, irrespective of
the actual used registers or whether an address register load was seen.

Additionally, since ntt switch all movs using the relative addressing are
actually used only once right on the next line, and are result of ntt converting

vec4 32 ssa_10 = intrinsic load_ubo_vec4 (ssa_0, ssa_9) (access=0, base=11, component=0)

into

  5: ARL ADDR[0].x, TEMP[0].xxxx
  6: MOV TEMP[2], CONST[0][ADDR[0].x+11]

RV530 shader-db:
total instructions in shared programs: 132966 -> 131904 (-0.80%)
instructions in affected programs: 29896 -> 28834 (-3.55%)
helped: 234
HURT: 2
total temps in shared programs: 16969 -> 16905 (-0.38%)
temps in affected programs: 604 -> 540 (-10.60%)
helped: 68
HURT: 12

Partial fix for: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7723

Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20577>

aco: Use NSA on GFX11 with more than 5 vaddr registers.

On GFX11 the first 4 vaddr are single registers and the last contains the remaining vector.
image_bvh64_intersect_ray has a special NSA layout.

Foz-DB GFX1100:
Totals from 2763 (2.05% of 134913) affected shaders:
VGPRs: 145884 -> 145056 (-0.57%); split: -1.03%, +0.46%
CodeSize: 18406864 -> 18326136 (-0.44%); split: -0.47%, +0.04%
MaxWaves: 76030 -> 76146 (+0.15%)
Instrs: 3559785 -> 3525287 (-0.97%); split: -0.97%, +0.00%
Latency: 44278460 -> 43303419 (-2.20%); split: -2.33%, +0.13%
InvThroughput: 4966295 -> 4914927 (-1.03%); split: -1.04%, +0.01%
VClause: 51755 -> 51991 (+0.46%); split: -0.05%, +0.50%
SClause: 105241 -> 105267 (+0.02%); split: -0.08%, +0.10%
Copies: 214141 -> 182419 (-14.81%); split: -14.82%, +0.01%
Branches: 69525 -> 69521 (-0.01%)
PreVGPRs: 120910 -> 120256 (-0.54%); split: -0.56%, +0.02%

No changes on Navi21.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20370>

aco: Validate GFX11 NSA correctly.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20370>

aco: Handle NSA with vectors in get_mimg_nsa_dwords.

No Foz-DB changes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20370>

mesa: add missing count_scale attributes

The EXT_gpu_program_parameters spec says:

   <params> points to an array of 4*<count> values

for both functions.

Cc: mesa-stable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20603>

mesa: add missing count_scale attribute

The EXT_gpu_program_parameters spec says:

   <params> points to an array of 4*<count> values

Fixes: a4e935f2d70 ("mesa: add EXT_dsa + EXT_gpu_program_parameters functions")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7943
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8025
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20603>

zink: lower stipple + smooth

We can use the new functionality in the draw-helper to implement
stippled smooth lines instead of what we currently do, which is aliased
stipping on smooth lines.

Reviewed-by: Soroush Kashani <soroush.kashani@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20134>

gallium/draw: support lowering stipple + smooth

When computing line smoothing, we can also do something similar to
compute the line stippling. This can be useful for some drivers, who
can't easily split the lines before rasterizing them.

This does lead to slightly inaccurate stippling, because the
line-smoothing extends the line-length by a small amount. That leads to
the line-stippling pattern being over-stretched over the line-segment by
a fraction of a pixel in lenght. For short lines, this can be quite a
lot of error.

Reviewed-by: Soroush Kashani <soroush.kashani@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20134>

egl: const display & config pointers in eglGetConfigAttrib()

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20567>

egl: inline _eglReportError(), and drop this and all the other unused defines

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20565>

radeonsi: rewrite si_update_ps_colorbuf0_slot to fix crashes and recursions

I'm convinced that u_blitter interactions with fbfetch can't be handled
in si_update_ps_colorbuf0_slot alone, so it has to be force-disabled
by si_blitter_begin. Another reason why it has to be disabled for u_blitter
and not ignored is because FBFETCH with MSAA enables sample shading
regardless of context states, and we don't want that for u_blitter.

Also, si_update_ps_colorbuf0_slot now disables FBFETCH explicitly before
its own DCC and CMASK decompression because even though u_blitter can't do
anything (due to blitter_running), si_blitter_end calls it too.

The result is that no recursion can occur thanks to the blitter_running
and suppress_update_ps_colorbuf0_slot flags, and FBFETCH is always
force-disabled before those flags are set, which is the state we want
to be in.

Fixes: bc6d22b9200 ("radeonsi: fix ps_uses_fbfetch value")

Acked-by: Yogesh Mohan Marimuthu <yogesh.mohanmarimuthu@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20318>

ci/lava: Tweak LAVA jobs timeouts

The Mesa CI LAVA job submitter was suffering from a bug in the LAVA
software that made their timeouts related to sub-actions unreliable,
such as waiting for the user login prompt automatic response.

The following MR
https://git.lavasoftware.org/lava/lava/-/merge_requests/1900 fixed this
issue. So we can now better control job timeouts granularity, failing
the job faster when there is something weird hanging the boot stage.

Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20596>

ci/lava: Update lavacli version

- Use new YAML loader derived from ruamel.yaml
- Remove PyYAML dependency from LAVA job submitter package

Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20596>

ci/venus: temporarily disable due to instability

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20622>

etnaviv: hide TS sharing behind debug option

TS sharing isn't fully stable yet. There are some fixes pending, but they
don't take care of all reported issues. Hide TS sharing behind a debug
switch until all the known issues are resolved.

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/20606>

docs: Add calendar entries for 23.0 release candidates.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20619>

broadcom/ci: drop duplicate job definition

See line 48 above.

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20605>

aco/gfx11: increase vgpr_limit to 256

fossil-db (gfx1100):
Totals from 280 (0.21% of 134574) affected shaders:
MaxWaves: 3124 -> 2846 (-8.90%); split: +3.46%, -12.36%
Instrs: 1139038 -> 1091407 (-4.18%); split: -4.18%, +0.00%
CodeSize: 5809332 -> 5486812 (-5.55%); split: -5.55%, +0.00%
VGPRs: 35004 -> 42864 (+22.45%); split: -1.85%, +24.31%
SpillSGPRs: 1896 -> 1865 (-1.64%); split: -2.37%, +0.74%
SpillVGPRs: 17807 -> 2382 (-86.62%)
Scratch: 2573312 -> 736256 (-71.39%)
Latency: 27470485 -> 17981296 (-34.54%); split: -34.54%, +0.00%
InvThroughput: 5606102 -> 6527051 (+16.43%); split: -4.19%, +20.61%
VClause: 32319 -> 19927 (-38.34%); split: -39.13%, +0.78%
SClause: 15014 -> 14897 (-0.78%); split: -0.95%, +0.17%
Copies: 102977 -> 93511 (-9.19%); split: -9.93%, +0.74%
Branches: 15164 -> 14969 (-1.29%)
PreSGPRs: 19132 -> 19014 (-0.62%)
PreVGPRs: 30494 -> 37460 (+22.84%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20251>

aco/gfx11: allow true 16-bit instructions to access v128+

It looks like the LLVM assembler promotes true 16-bit instructions to VOP3
in this case.

No fossil-db changes.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20251>

aco: disallow SGPRS/constants with interpolation instructions

https://reviews.llvm.org/D137575

The VINTRP format cannot encode anything except VGPRs.

Reading VINTERPInstructions.td, looks like it's the same for GFX11.

No fossil-db changes.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20251>

aco: add more opcodes to can_use_DPP()

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20251>

aco: allow Builder::Result to be dereferenced

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20251>

aco/gfx11: fix discard early exit removal optimization

This optimization never happened because the NULL target was removed in
GFX11.

fossil-db (gfx1100):
Totals from 5439 (4.04% of 134574) affected shaders:
Instrs: 407865 -> 387123 (-5.09%)
CodeSize: 2163340 -> 2060644 (-4.75%)
Latency: 3432378 -> 3327802 (-3.05%)
InvThroughput: 270133 -> 262980 (-2.65%)
Branches: 8524 -> 3085 (-63.81%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20513>

microsoft/spirv_to_dxil: Let linking report the need of runtime data

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20608>

etnaviv: properly check for new clear value in sampler TS setup

Comparing the 64bit clear value to the lower half 32bit clear state is
obviously wrong and results in a lot of false positives.

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/20586>

etnaviv: blt: properly use upper half of clear value

Instead of replicating the lower half of the clear value, properly
use the upper half to program the second clear value BLT state.

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/20586>

ir3/print: Consider the size of input when matching it to a reg

First component of the input may never be read, so we have to find
the first component of the input which is being read.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20533>

ir3: Force flat.b sources to be equal

From cc29b940 according to Jonathan Marek using 2 immediates in a cat2
instruction is only "correct" if they are both equal.

They were not equal after reassigning of packed inlocks.

flat.b seemed to work fine even with imm values being different,
but better not to risk.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20533>

tu: Fix varyings interpolation reading stale values

tu6_vpc_varying_mode returned how many bits are consumed IF they are
non-zero, for SMOOTH mode nothing is written and it was treated like
no bits were consumed.

When input with smooth interpolation was the last one and straddled
the VPC_VARYING_INTERP_MODE regs, the last interp mode were not written.

Fixes misrendering in "Psychonauts 2".

Cc: mesa-stable
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20533>

loader/dri3: Overhaul draw->cur_num_back handling

draw->cur_num_back is no longer used for immediately reducing the number
of allocated back buffers. It just reflects how many of them are
currently allocated, and is used to prevent allocating more if the
current maximum is already reached.

This fixes an issue where the current in-progress back buffer could get
freed, which would result in visible artifacts.

Unused buffers are freed once they exceed buffer age 200. This ensures
we do not keep around more buffers than necessary in the long run.

v2:
* Drop buffer age threshold to 200. This will free unused buffers more
  quickly, and seems enough to avoid doing so prematurely even with
  "glxgears -fullscreen" running at thousands of frames per second.
* Use "buf_id != LOADER_DRI3_FRONT_ID". (Pierre-Eric Pelloux-Prayer)
* Simplify loader_dri3_get_buffers changes slightly.
* We can now use any unallocated back buffer slot in dri3_find_back.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20385>

loader/dri3: Add dri3_set_render_buffer helper

Preparation for next commit, no functional change intended.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20385>

loader/dri3: Move NULL handling into dri3_free_render_buffer

Both checking for draw->buffers[buf_id] being non-NULL, and setting
draw->buffers[buf_id] = NULL.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20385>