radv: fix binding raytracing/compute pipelines

If a compute pipeline is bound after a raytracing pipeline, the
computes shader slot (aka RT prolog) will be overwritten.

To fix this, move the RT prolog outside of the compute shader slot.

Fixes: d109362a3da ("radv: copy bound shaders to the cmdbuf state")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22235>

radv: add the raygen shader BO to the cmdbuf list

Found by inspection.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22235>

ac/nir/ngg: Slightly improve attribute ring offset calculation.

Inspired by Nicolai Hähnle's commit in LLPC.
Instead of using a SALU instruction to add to the scalar
offset, rely on the buffer swizzling and use constant offset.

Fossil DB stats on GFX1100:

Totals from 47910 (35.51% of 134913) affected shaders:
CodeSize: 87927612 -> 86968136 (-1.09%)
Instrs: 17584007 -> 17440094 (-0.82%)
Latency: 97232173 -> 97126311 (-0.11%)
InvThroughput: 9904586 -> 9905288 (+0.01%); split: -0.02%, +0.02%
VClause: 544430 -> 542566 (-0.34%)

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22227>

radv: Use radv_get_shader to get vertex shader when binding pipeline.

The shaders[MESA_SHADER_VERTEX] can be NULL for merged shaders.

Fixes: b2ac40e734194fcb571a72396b5644b1d351e48a
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8749
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22229>

radv: configure PA_SC_MODE_CNTL_1 during cmdbuf recording

Two graphics pipeline parameters need to be copied to the cmdbuf
state.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22218>

radv: set PS_ITER_SAMPLE(1) for sample shading during cmdbuf recording

This shouldn't be configured in the pipeline.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22218>

radv: copy db_render_control to the cmdbuf state

This register is only used for meta operations.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22218>

iris: Implement Xe version of bo_madvise() and bo_set_caching()

Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22060>

iris: Place scanout buffers only into lmem for discrete GPUs

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22060>

iris: Handle allocation of scanout buffers in Xe

Bos that will be scanout in display need to be allocated with
flags = XE_GEM_CREATE_FLAG_SCANOUT in Xe and that implies to different
caching rules for this buffer.

So here not allowing to get scanout buffer from cache or allow it
to be placed in a cache bucket for reuse.

Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22060>

iris: Handle allocation of exported buffers in Xe kmd

Bos that will be exported need to be allocated with vm_id = 0 in Xe,
so don't try to get a bo from cache that was allocated with a
valid vm_id.

Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22060>

iris: Add BO_ALLOC_SHARED

Xe KMD requires special handling for exported buffers during creation.

Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22060>

anv: Use the new vk_device_memory base struct

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

vulkan: Add a vk_device_memory base struct

This lets us provide a vk_device_memory_range helper similar to what's
provided for buffers for dealing with VK_WHOLE_SIZE.  We can also handle
flags and some annoyance around Android hardware buffer import.

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

vulkan,anv,hasvk,radv: Unify Android hardware buffer creation

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

radv: Set vk_image.ahardware_buffer_format

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

anv,hasvk: Set vk_image.ahardware_buffer_format

This effectively just modifies it for YCbCr images where the common code
doesn't have a useable format mapping.

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

vulkan: Add an ahardware_buffer_format field to vk_image

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

vulkan/android: Fix hardware buffer usage flags

We now add the correct usage flags for input attachments, storage
images, and depth/stencil attachments.

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

vulkan,anv,hasvk,radv: Add a common vk_image_usage_to_ahb_usage helper

Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>

anv: set CFE_STATE.OverDispatchControl to default

BSpec specifies default value for CFE_STATE.OverDispatchControl
is 2, or 50% overdispatch. No observed performance impact.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22179>

anv: only emit CFE_STATE when scratch space increases

On Gen12.5+, we only need to emit CFE_STATE when scratch space
has changed, not on every pipeline binding. Also, only grow the
scratch space, never shrink it. Need to reset after secondary buf.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22179>

anv: move queue check helpers to anv_private

Also fix missing trace point stuff in command buffer begin/end

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Felix DeGrood <felix.j.degrood@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22179>

anv: compute the largest GRL kernel scratch size

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Felix DeGrood <felix.j.degrood@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22179>

anv: cs_stall during compute state flush on < gen12.5

The CS Stall in cmd_buffer_flush_compute_state is only required
on HW that uses MEDIA_VFE_STATE, gen12 and previous.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22179>

gallium: Modify default path for DMABUF to use DRM

Modify the code path taken in `u_pipe_screen_get_param_defaults`
to call DRM to check if `PIPE_CAP_DMABUF` is supported. This is
required for overriding the behavior in `dri2_init_screen_extensions`
to support importing DMA bufs on drivers that don't support DRM, by
simply changing how `PIPE_CAP_DMABUF` is handled in their driver.

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

d3d12/llvmpipe/softpipe: Add support for `get_screen_fd`

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

r300/r600/radeon_si: Add support for `get_screen_fd`

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

virgl: Add support for `get_screen_fd`

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

svga: Add support for `get_screen_fd`

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

i915: Add support for `get_screen_fd`

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

iris: Add support for `get_screen_fd`

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

zink: Add support for `get_screen_fd`

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

nouveau: Add support for `get_screen_fd`

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

tegra: Add support for `get_screen_fd`

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

crocus: Add support for `get_screen_fd`

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>

lavapipe: advertise EXT_shader_object

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22233>

lavapipe: implement EXT_shader_object

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22233>

lavapipe: break out pipeline layout creation for reuse

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22233>

venus/ci: Only run one crosvm instance

As venus now has separate render server processes it is better to use a single
crosvm instance to run all the tests.

Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21693>

ci: Uprev crosvm and virglrenderer

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21693>

vulkan: Update XML and headers to 1.3.246

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22231>

zink: update requirements now that pv mode can be emulated

Zink can now emulate provoking vertex mode when the extention is
missing, update requirements to reflect this.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22162>

zink: always advertize provoking vertex mode support

Require VK_EXT_provoking_vertex for optimal_keys and always advertize
PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION since it can now be
emulated

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22162>

zink: add provoking vertex mode lowering

Can be used as fallback for when VK_EXT_provoking_vertex is missing

Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22162>

zink: add field to 'zink_gs_key' and enum

Add enum for pv emulation primitives and `lower_pv_mode`
to `zink_gs_key`

The enum contains the possible values of the lower_pv_mode key

This key will be non 0 whenever provoking vertex mode needs to be
emulated and it's exact value encodes relevant information about the
primitive that needs to be emulated

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22162>

zink: simplify logic to call `zink_set_primitive_emulation_keys`

The logic had grown to check every primitive indivdually, instead just
check wehether `rast_prim` has changed

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22162>

nir: only handle flat interpolation when needed in `nir_create_passthrough_gs`

When turning primitives into line strips this function needs to move
attributes around, but this is not needed in other cases.

Fixes: 1a5bdca2dda ("zink: implement flat shading using inlined uniforms")
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22162>

panfrost: Check the return value of drmGetVersion

Signed-off-by: xurui <xurui@kylinos.cn>
Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Reviewed-by: Italo Nicola <italonicola@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22203>

docs: make code-block indents consistent

Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22052>

docs: format code-block as toml

Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22052>

docs: format code-block as ini

Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22052>

radv: Work around use-after-free compiler errors

Fixes: 4dafb69d ("radv/rt: defer library_pipeline allocation")
Tested-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21981>

radv: add dynamic support for rectangles enable/mode

This is in VK_EXT_discard_rectangles version 2.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21360>

vulkan: add dynamic support for rectangles enable/mode

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21360>

radv: fix sample shading when a new fragment shader is bound

Sample shading can be enabled from the pipeline and from the fragment
shader, in that case we need to re-emit some states.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22215>

intel/decoder: Bump the binding table guess value to 32

Let's bump the binding table value guess to 32 instead of 8 and also stop
decoding the binding table entries as soon as we hit the zero
initialized value in the map.

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22221>

docs: Remove docs about macOS hardware drivers

We dropped support for hardware macOS drivers in afe134a49c5 ("asahi: Drop macOS
backend"), so drop the corresponding documentation. Layered and software drivers
are still supported on macOS for better or worse, so the main "Notes on macOS"
page can stay I think.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22213>

microsoft/clc: Add shader model / validator to compiler API

Shader model 6.2 was the upper bounds of what *could* be generated
before, but not all devices support it. And other devices support
even more. So, let's pass in the shader model / validator that will
be used by the API caller.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21178>

panfrost: Choose hierarchy masks by vertex count

Currently, we always use a hierarchy mask with all levels enabled. While this is
efficient for geometry-heavy workloads like 3D games, it is wasteful for 2D
applications that draw very few vertices. For drawing just a few textured quads,
the overhead of small bin sizes outweighs any performance advantages, so it's a
bit slower. More problematically, small bin sizes require tremendous amounts of
memory for the polygon lists, leading to significant memory consumption (~10MB)
for the polygon list for even the simplest of 2D blits.

To reduce our memory footprint, we need to choose our hierarchy masks more
carefully. In general, we want to allow small bin sizes for geometry-heavy
workloads but not for geometry-light workloads. We estimate vertex count in the
driver as a proxy for this, and use a simple heuristic to select a bin size
based on the estimated vertex count. None of this is an exact science, and the
heuristic could probably be tuned. Nevertheless, the heuristic used (comparing
framebuffer size to vertex count) works well in practice, significantly reducing
the memory footprint of 2D applications like Firefox without hurting the
performance of 3D applications.

I originally wrote this patch while diagnosing high memory footprints on my
Midgard laptop, which is why only Midgard is in scope here. On Bifrost and
Valhall, we have a similar hiearchy mask selection problem. It seems likely that
the same heuristic would work there too, but it's a different code path that I
have not integrated or tested. I'll leave that for the adventurous reader, to
get the memory footprint win there too.

(It's also possible the win is smaller on newer Malis than on Midgard, since Arm
claims they optimized the tiler data structures on the newer parts. There's
probably still some merit to the idea.)

On Mali-T860, glmark2 -bdesktop frametime decreased by 1.35% +/- 0.91% at 95%
confidence, showing a slight win for 2D workloads No statistically significant
difference for glmark2 -bshading:shading=phong, since 3D workloads continue to
use the same hierarchy masks.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19482>

panfrost: Estimate vertex count for hier mask

In the next commit, we will refine our algorithm to select hierarchy masks based
on the vertex count. In preparation, augment the driver to track rough estimates
of the vertex count so we have a "geometry complexity" input for the heuristic.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19482>

panfrost: Clean up tiler calculations

We're about to do some work on this file. Clean it up first.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19482>

freedreno/computerator: Add support for a7xx

Not everything works correctly, e.g. stib seems flakey while stg
seems alright.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno/computerator: Templatize a6xx backend

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno: Add dummy a730/a740 definition

Needed for assembly/disassembly.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno: Move fd6_pack.h to common code accessible by computerator

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno/computerator: Convert to C++

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno: C++ fixes for computerator to compile

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno/computerator: C++ proofing

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno/registers: More a7xx regs

Based on 011c54b0 from Jonathan Marek.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

freedreno/registers: Document new CP_EVENT_WRITE::SEQNO

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22148>

panfrost: Add a v9 fast path for no images

The usual case.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21848>

panfrost: Don't update access with a single batch

drawoverhead test 25 from 462->492

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21848>

panfrost: Mark packs as ALWAYS_INLINE

As Intel does. These functions are written with the expectation that they will
be inlined away, allowing gcc's copy-prop and constant folding to eliminate the
template struct and any unused fields.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21848>

panfrost: Don't redundantly call emit_const_buf

On Valhall, we were calling emit_const_buf in two places:

1. The main "handle dirty flags" code shared with Bifrost
2. A Valhall-specific shader environment emitter

The latter was not dirty tracked, and the former was not used. That meant we
were calling emit_const_buf way too much. It's not a cheap routine, either.

Instead, use the results from the dirty tracked function in the shader
environment emitter, to avoid the redundant call and get the expected dirty
tracking.

In a Dolphin trace I'm looking at, fps increases 27->33.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21848>

panfrost: Print perf debug on seqnum overflow

Another unexpected source of flushes.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21830>

panfrost: Print perf debug when flushing everything

..Even if the only batch is the one that's currently bound.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21830>

zink: stop caching vertex states

I tried to be too clever and ended up wasting cpu cycles. it's
much, much, much, much faster to just generate this one struct array
every time than it is to do set lookups with thousands of members

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: use fast popcnt for vstate draws

also delete some unused stubs for no dynamic vertex input since I'm never
gonna implement that path

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: don't swizzle velems state for vstate draws

this isn't ever used, so don't touch it

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: explicitly pass null velems when creating pipelines with dynamic vinput

this may or may not be a usable pointer, and it's not being read, so
don't pass it at all

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: add another vstate draw template for popcnt presence

matching radeonsi

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: bind vertex state directly from draw hook

this is more streamlined and readable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: use search_or_add for masking vstate

this should be a significant perf boost instead of multiple lookups

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: flag vertex buffers for rebind after vstate draws

vstate draws bind their own vertex buffers unrelated to the bound
gallium buffers, so any draw occurring after a vstate draw must
rebind vertex buffers to ensure the correct ones are bound

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

zink: omit VkPipelineVertexInputStateCreateInfo with dynamic vinput

this should never be used/needed

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>

radv: Stop counting user SGPRS separately

Renames radv_declare_shader_args to declare_shader_args and runs it
twice to first gather the user SGPR count without push constants and
descriptor sets.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22119>

radv: Set user SGPR locations when declaring args

Merge shader arg declaration with setting up the user data locations.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22119>

bin: Fix typos

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22189>

zink: block resolves where src extents > dst extents

vulkan resolves only provide "extents" instead of src and dst regions like
GL, which means vk resolves can't be used to downscale images, as such
operations will instead just crop the image

fixes #8655

cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22195>

radv: add push constant state to the cmdbuf state

Push constants are handled per bind point internally. Using a separate
structure in the cmdbuf state would allow us to update it easily
without relying on bound pipelines.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22209>

radv: copy need_indirect_descriptor_sets to radv_cmd_state

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22209>

radv: add a helper to convert a VkPipelineBindPoint

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22209>

freedreno: Optimize repeated finishes

Sometimes apps (glances at stk) spin on a syncobj with very short
timeouts.  But ensuring the fence is flushed all the way through to
the kernel (including handling TC unflushed fences) only needs to
be done once.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>

freedreno/drm: Disable threaded-submit for msm

We've had drm/sched support on the kernel side for more than a year and
a half.  This makes submit ioctl async by handling fence waits from the
sched's kthread, which is what threaded submit was originally working
around.  For now, threaded submit is only used for virtgpu, which does
not (yet?) have drm/sched support.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>

freedreno/drm: Make threaded-submit optional

We've had gpu-sched support in the kernel for a while now, so our fence
waits are not synchronous in the ioctl path.  The only reason this path
still exists is that virtgpu does not have gpu-sched.  So lets disable
it on msm.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>

mesa: Add a few more function traces

Sprinkle around a few more traces that were useful in locating fence
waits.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>

freedreno/drm: Stop cleanup at first active BO

Buffers are added to the deferred freelist at the tail.  And frequently
the last reference is dropped immediately after the submit.  So almost
always, once we see a still-busy BO, the remaining in the list will also
still be busy.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>

freedreno/drm: Fast path for idle check

If already idle, no need to cleanup_fences() (and take related lock).

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>

zink: fix quads emulation gs with array variables

this was broken for e.g., gl_ClipDistance, which uses explicit array
types and therefore cannot be directly read/written

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22208>