panvk: Report row_stride in GetImageSubresourceLayout

...Rather than line_stride. For linear images, these are equivalent. For
nonlinear images, rowPitch is implementation-defined. So this isn't strictly a
bug fix, but it gets rid of the nonsense nonlinear line_stride.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16201>

panvk: Remove unused layout structs

PanVK switched to the common Panfrost layout code, but these duplicate structs
stuck around. Garbage collect them to prevent confusion.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16201>

pan/decode: Print row strides, not line strides

I.e. what's actually passed to the hardware, in case of compression or tiling.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16201>

pvr: Add compute update kernel in vkCmdDispatch().

We upload a new data section whenever a patched variant is
needed. They will be freed at command buffer destruction since
the uploads are linked.

Co-authored-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16040>

pvr: Add compute update shared kernel in vkCmdDispatch().

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16040>

pvr: Add initial implementation of vkCmdDispatch().

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16040>

pvr: Add PVR_WORKGROUP_DIMENSIONS define.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16040>

anv: fix invalid utrace memcpy l3 config on gfx < 11

device->l3_config is only valid on Gfx11+

This only fixes using GPU_TRACE=1

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 02a4d622ed0 ("anv: expose a couple of emit helper to build utrace buffer copies")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16291>

llvmpipe: disable compute statistics queries if queries are disabled

cc: mesa-stable

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16278>

llvmpipe: remove misleading query comment

u_blitter already handles this

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16278>

llvmpipe: disable statistic queries during draw if queries are disabled

cc: mesa-stable

fixes (zink):
spec@ext_transform_feedback@generatemipmap prims_generated

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16278>

radeonsi: generate unique shader name in si_get_nir_shader

This function modifies the NIR shader, so when using NIR_DEBUG=print and
nir_viewer all the variants are displayed using the original name.

With this change, we get the original shader (eg: GLSL3), and then each
variant gets its own name (GLSL3-xxxxxxxx) - and is displayed in its
own tab in nir_viewer.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16247>

radeonsi: don't always strip NIR debug infos

When using NIR_DEBUG=print, always having shader names help debugging.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16247>

aco: Use movk for AddressHi bits in vertex prolog

Eliminates a useless extra dword by transforming
        s_mov_b32 s47, 0xffff8000                                   ; beaf03ff ffff8000
to
        s_movk_i32 s47, 0x8000                                      ; b02f8000

which does the same thing.

Reviewed-By: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15415>

aco/ra: fix live-range splits of phi definitions

It could happen that at the time of a live-range split,
a phi was not yet placed in the new instruction vector,
and thus, instead of renamed, a new phi was created.

Fixes: dEQP-VK.subgroups.ballot_broadcast.compute.subgroupbroadcast_i8vec2
Cc: mesa-stable
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16248>

ac/nir: Do not set cursor in try_extract_additions

Fixes: 61ac5ac ("radv,ac/nir: lower global access to _amd global access intrinsics")
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16270>

radv: use 3D views for 3D internal operations on GFX6-8

Instead of binding 2D images views and relying on BASE_ARRAY.
This cleanups the internal driver operations for 3D.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16272>

nir: fix sorting before assigning varying driver locations

We need to make sure we also properly sort varyings sharing a single
slot otherwise we can end up assigning earlier components to the next
slot if we have already processed later components.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6392

Fixes: 1e93b0caa10d ("mesa/st: add support for NIR as possible driver IR")

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16208>

util: Add some unit tests of the half-float conversions.

Make sure that we're handling nans/infs correctly, in particular only
generating quiet nans when given a quiet nan.  This catches the previous
qNaN fix.

Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16233>

util: Keep quiet NaNs quiet when converting to half float.

We don't want to be throwing exceptions and changing float values later by
emitting a signaling binary16 nan.

If we don't do this, then when we convert back to f32 in NIR constant
expression evaluation, the signaling NaN can end up giving NaN for
fmax(NaN, 0.0), instead of 0.0.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5933
Cc: mesa-stable
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16233>

iris: use new kernel uAPI to compute video memory

v2: Use os_get_available_system_memory() when kernel memory region
    uAPI is not available (Lionel)

Cc: 22.1 <mesa-stable>
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16210>

panvk: Add address/range helpers for panvk_buffer

The range helper is taken from ANV; the gpu_ptr one is original.  This
also fixes a few more bugs where we weren't adding offsets in properly.

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16216>

panvk: Only write depth when depth test enabled

Fixes dEQP-VK.pipeline.depth.format.*_test_disabled.depth_write_enabled

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16283>

panvk: Lower image deref atomics

Fixes dEQP-VK.image.atomic_operations.*

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16283>

pan/bi: Handle texture offset + index

Fixes dEQP-VK.glsl.opaque_type_indexing.sampler.uniform.vertex.sampler1d

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16283>

panvk: Allocate stack for each thread

Rather than a single stack for all threads to share -- that can't work! :-) We
use the same helper that the GLES driver does. Fixes anything using scratch or
spilling, including:

dEQP-VK.glsl.indexing.varying_array.vec3_static_write_dynamic_read

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16283>

panvk: Lower global vars to local

Fixes dEQP-VK.glsl.indexing.varying_array.float_static_write_static_loop_read

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16283>

venus: log performance envs upon instance creation with debug init

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Change-Id: I8db93d6de28ac6a727c10fbaca94cffac092f8ea
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16256>

venus: add VN_RELAX_BASE_SLEEP_US env var for tuning cpu utils

Cap the minimum to the current base sleep us 10.

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16256>

drm-shim: Cleanup on device file close

Avoid leaking memory when the device file is closed, dropping bo
references and freeing the shim_fd.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>

freedreno/drm-shim: Better iova handling

We actually want to use util_vma to handle this.  But fortunately core
drm-shim alredy does this for mem offset, we can just delete a bunch of
code and re-use that.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>

drm-shim: Add error handling for bo_init()

Don't assert, return an error.  If we are fuzzing something using
drm-shim, we want to be more like the kernel and return an error
for OoM situation, rather than falling over.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>

drm-shim: Better mmap offsets

Using the bo pointer address as the offset doesn't go over well when
someone is fuzzing you.  But we already have the mem_addr, we can simply
use that instead.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>

freedreno/drm-shim: Robustify error handling

We can't be so sloppy if we are using drm-shim for fuzzing.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>

freedreno/drm-shim: Update to latest uapi version

Needed for fuzzing virgl drm native context.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>

drm-shim: Add GET_UNIQUE support

Needed by drmOpenWithType(), which is used by virgl drm native context.
(We want to use drm-shim for fuzzing.)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>

vulkan/wsi/wayland: Fix double free on error condition

If wsi_configure_native_image() fails, it will call
wsi_destroy_image_info() itself, so let's try to not call it again from
wsi_wl_swapchain_destroy().

Fixes the CTS tests:
dEQP-VK.wsi.wayland.swapchain.simulate_oom.*

Fixes: b626a5be43d ("vulkan/wsi/wayland: Split image creation")

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16257>

vulkan/runtime: 0-out pipeline cache object

I'm running into crashes because cache->cache_object is uninitialized.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 591da9877900 ("vulkan: Add a common VkPipelineCache implementation")
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16275>

intel/compiler: In XeHP prefer <1;1,0> regions before compacting

Ken performed some tests with shader-db to evaluate the effects

```
Across all 145,848 shaders generated, the results were:

Total bytes compacted before: 3,326,224
Total bytes compacted after: 60,963,280
```

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15399>

llvmpipe: add a ci flake

not able to repro locally

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16280>

d3d12: Destroy empty residency bo set during early-return

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

d3d12: Fix a couple over-releases from incorrect take-ownership flag

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

d3d12: Fix leaks in map with do-not-wait

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

d3d12: Use a pipe_reference in d3d12_bo

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

u_primconvert: Handle take_index_buffer_ownership

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

u_primconvert: Refactor to remove recursion from util_primconvert_draw_vbo

This prevents having unused copies of pipe_draw_info and
pipe_draw_start_count_bias on the stack, and makes it easier to do
things once for a multi-draw, which will matter in the next patch.

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

u_debug_stack: Lock around stack dumps to prevent interleaving

Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

u_debug_symbol: Allow re-init of DbgHelp to work around already-initialized issues

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

u_debug_symbol: Use correct sizeof for DbgHelp

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

u_debug_refcnt: Don't loop for initial refcounts if the initial value is huge

Some components use very large refcounts to bypass atomics in single-threaded
cases, and this produces near-infinite loops when these resources are created.

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

bin: Add a script for filtering/sorting refcount logs

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16182>

mesa: Ensure ARB programs end in a newline

If the last line of an ARB program has a comment, the program will
fail to parse, because the lexer only considers a comment valid if
it ends in a newline, not EOF. The parser then fails on the '#'.

Reviewed-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16230>

d3d12: Correctly key off of polygon stipple enable cap

Apparently we were only keying off the presence of a real stipple pattern
being set, and completely ignoring when the app does glDisable().

Add in the actual enable bit as an additional discriminator to determine
if we should be doing polygon stippling.

Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16254>

pan/va: Add non-IDVS VAR_TEX instructions

Used in blit shaders.

Icecream95 supplied the test cases.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Add VAR_TEX_BUF_GRADIENT instruction

Probably useful for ... something.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Improve texture instructions

[Alyssa: Related improvements.]

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Add absneg modifier to V2F32_TO_V2F16

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Add SEG_ADD/SEG_SUB operation

[Alyssa: Fixes to the instruction.]

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Add more transcendental operations

TODO: Make the secondary opcode field wider so that FATAN_ASSIST can
be split into two instructions

[Alyssa: Fixes to the hardware behaviour.]

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Improvements to LEA_TEX instructions

[Alyssa: Fixes to match hardware.]

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Mark NOT instructions as deprecated

The inverters on the bitwise ops should be used instead.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Add missing TABLE (SFU) instructions

Equivalent to their Bifrost predecessors.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

pan/va: Generalize message? check for asm

Allows passing more uniforms in more places. We'll use this in a test case in a
moment.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15588>

panfrost: Flip point coords in hardware

On Bifrost, this is very easy: there's an RSD bit to Y-flip gl_PointCoord. It
should map perfectly to the Gallium bit. With this change, we no longer use
lower_pntc_ytransform on Bifrost, saving a bit of ALU when reading point
coordinates.

On Valhall, this is quite hard: the bit is in the framebuffer descriptor now!
That means it can't be changed in a batch. This is expected to be ok: on GLES
and VK, the origin is controlled only by the framebuffer orientation. It's a
bigger problem on big GL, where GL_POINT_SPRITE_COORD_ORIGIN can be set freely.
To cope, a tri-state data structure is used for the state tracking. This has a
failure case on Valhall: every draw toggling the coord origin. However, the
intention of the ORIGIN state bit is smoothing over coordinate system
differences; it should never /actually/ change once set. Until we see an app
doing something so stupid, I don't think we should worry about.

We need all the Valhall tri-state infrastructure for handling provoking vertices
on big GL anyway.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Lower user clip planes

Since we don't export the relevant CAP, the state tracker calls
nir_lower_clip_vs for us. However, for some reason we're still responsible for
calling nir_lower_clip_fs. Now that we have sane shader key infrastructure,
let's do so.

Fixes the floor rendering wrong in the title screen of Neverball.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Lower point sprites on Bifrost

Use the common pass. This only should trigger when drawing points, so we need
some extra tracking to ensure this.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Refactor variant rebind code

For point sprite lowering on Bifrost and Valhall.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Refactor variant selection code

Extract the "compile a new variant" routine from the "select and bind a variant"
routine. This allows us to simplify the control flow, eliminating the `compiled`
boolean on the shader structure.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Simplify shader key architecture

Rather than clever key compare/initialize code, let's make the shader keys plain
old data. This makes it easier both to extend and to optimize the shader keys.

Keys are compared with a simple memcmp(). I considered a hash table but I don't
think we have enough variants (or large enough keys) to justify the overhead.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Use s->info.stage instead of tgsi stage

Now that we have the NIR handy.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Don't subclass pipe_compute_state

Just copy the bit we need so the NIR doesn't hang around. This is simpler.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Call tgsi_to_nir earlier

This ensures we always have NIR available in the shader state. It also saves a
(trivial) amount of recomputation if multiple TGSI variants are needed.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panfrost: Remove ancient unused code

Leftover scraps.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16173>

panvk: Emit fragment RSDs even with no shader

In Vulkan, it's possible to create a pipeline with no fragment shader that's
still expected to rasterize. This is useful for depth/stencil side effects, and
is closely related to the "fragment shader required" optimization we do in the
GLES driver. Refactor the RSD emit code to handle this case.

Fixes dEQP-VK.pipeline.stencil.nocolor.*

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16204>

panvk: Gate rasterization on !discard, not fs req

The "fragment shader required?" computed state is about fragment shader side
effects. There may be no fragment shader required but depth/stencil side effects
meaning that rasterization is nonoptional. What actually gates rasterization is
the rasterizer discard bit. Use that instead.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16204>

panvk: Streamline no shader RSD case

Noticed by inspection.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16204>

panfrost: Only flip point sprites for GL

Fixes dEQP-VK.glsl.builtin_var.simple.pointcoord

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16204>

panvk: Use correct point size limits

As determined by Icecream95. Fixes:

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

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16204>

panvk: Ignore point size for !points

Otherwise wide lines break. The alternative approach is to eliminate the points
writes when not drawing points since we do have topology information at compile
time. I'm admittedly stuck in my GL mindset. That's the approach we'll need for
Valhall anyway.

Fixes dEQP-VK.rasterization.interpolation.basic.lines_wide

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16204>

pan/bi: Use texture index instead of sampler for message preloading

The VAR_TEX definition in ISA.xml only has a field for texture_index,
so trying to read sampler_index will return zero; read from
texture_index instead, and rename other fields for consistency.

The texture and sampler indices must be equal for VAR_TEX to be used,
so either name could be used for the field.

Fixes the wrong textures being used in Thief.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6219
Fixes: eb1479bda22 ("pan/bi: Support message preloading")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16255>

r600: Allow eight bit, three channel formats for vertex buffers

While using three component texture formats results in CTs failures,
three component vertex attributes are fine, and not allowing them
results in significant performance regressisons.

Fixes: e41958e344cb4b15d01008140a1ee08817104334
   r600: Disable eight bit three channel formats

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6399

v2: rename function to is_buffer_format_supported (Emma)

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16267>

anv: limit clflush usage

Discrete platforms don't have LLC, but on those, we mmap our buffers
with WC. So we shouldn't need to clflush there.

Anv already had a boolean field on the physical device to know whether
we need to use clflush(), based off the memory heaps available. So use
that instead.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15780>

anv: fix clflush usage on utrace copy batch

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: cc5843a573bd ("anv: implement u_trace support")
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15780>

radeonsi: fix VS kill_outputs handling

981bd8cbe2d moved outputs removing handling to NIR, but instead of
applying it only to the last stage before the FS this now applies
it to both the GS and the VS.

This commit fixes this by clearing the kill_outputs field for
the VS when using a ES-GS shader.

Fixes: 981bd8cbe2d ("radeonsi: apply key.ge.opt.kill_{outputs,pointsize,clipdistance} in NIR")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16249>

radv: consider radix sort shaders to be internal/meta

Cleans up RADV_DEBUG=shaders

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16198>

v3dv: ignore barriers for image layout transitions from undefined layouts

Layout transitions are not relevant to us, we only care about barriers
that involve a sync point between read/write actions on the image across
GPU jobs.

Image transitions from undefined layout can only happen before the image
is ever used by the GPU, which means they are never relevant to our
implementation.

This improves performance in vkQuake.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16235>

v3dv: document Vulkan requirements for signaling operations

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16235>

v3dv: don't emit final noop job if we don't have anything to signal

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16235>

v3dv: drop obsolete comment

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16235>

v3dv: ignore no-op barriers

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16235>

agx: Lower mediump flat shading

This isn't supported by the hardware. Fixes

dEQP-GLES2.functional.shaders.constants.float_uniform_vertex

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Remove nir_register support

We don't use it anymore, now that we can handle SSA form. Gets rid of
the most gross hack in the compiler.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Remove has_liveness

Given we do no metadata tracking, this is an accident waiting to happen.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Update RA comment

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Add validation pass

For now, just check that we didn't botch the structure of the block,
since this just bit me.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Remove identity moves

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Unit test parallel copy lowering

It's pretty tricky.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Always use hash table for extracts

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Split up RA from post-RA lowering

This allows us to validate results in the middle.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

agx: Lower phi nodes to parallel copies

Now we have an SSA RA :-)

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>