nvc0: support PIPE_CAP_RESOURCE_FROM_USER_MEMORY_COMPUTE_ONLY

v2: rework by adding a new buffer status

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4580>

gallium: add PIPE_CAP_RESOURCE_FROM_USER_MEMORY_COMPUTE_ONLY

With the current UAPI we only support user pointers from the compute
engines, so we need a way to express that in gallium.

v2: fix typos
v3: add allows_user_pointers helper

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4580>

nouveau: enable HMM

v2: move declarations into libdrm
v3: fix typos
    rework handling of how much memory to reserve
v4: remove unused parameter
    unmap cutout on error and when the screen is destroyed
v5: move into screen_create
    enable HMM only if CL gets enabled

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4580>

ci: bump libdrm to 2.4.102

Since version 2.4.101 there are only xz archives hence the bz2 to xz
change.

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4580>

zink: block resolve blits for depth/stencil buffers

"The format features of dstImage must contain VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT"
- vkCmdResolveImage spec

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5888>

zink: block vkCmdBlitImage usage for multi sampled blits

this is prohibited by spec

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5888>

zink: try copy_region hook for blits where we can't do a regular blit or resolve

in cases where the formats match, we can likely just pass this through for now

fixes a ton of spec@!opengl 1.1@depthstencil-default_fb-blit piglit tests

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5888>

mesa/main: use call_once instead of open-coding

We already have a utility for this, so let's use that instead.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5879>

mesa/main: factor out one-time-init into a helper

This will make the next commit a bit cleaner.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5879>

gv100/ir: set ftz bit on floating point operations

Fixes Unigine Heavens ambient occlusion

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5881>

meson: fix android vulkan build

Android doesn't have `pthread_cancel()` and is unlikely to ever
implement it [1], but `wsi_common_display.c` needs it (or an
alternative).

Let's just disable the platform on Android (as it used to be
before 448eb19158f483d807ef).

[1] https://android-review.googlesource.com/c/platform/bionic/+/1215779/1/docs/status.md

Fixes: 448eb19158f483d807ef ("vulkan: automatically compile the `display` platform when available")
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Acked-by: Nataraj Deshpande <nataraj.deshpande@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5860>

gitlab-ci: Drop dependencies:

Artifacts from jobs listed in needs: are downloaded by default, so
there's no need to list them in dependencies: as well (in fact,
https://docs.gitlab.com/ce/ci/yaml/#artifact-downloads-with-needs says
using dependencies: together with needs: is invalid, so we might have
been getting lucky...).

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5845>

gitlab-ci: Remove indirect dependencies from needs:

Tomeu discovered that GitLab 12.8 fixed the bug where jobs would
spuriously run even though some of their dependency jobs were skipped.
So we don't need to list indirect dependencies anymore.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5845>

tu: Don't invalidate irrelevant state when changing pipeline

At least in the future this could let us avoid re-emitting gfx/cs
constants when the other changes. This also matches what the blob does.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5877>

freedreno/a6xx: Add some documentation for shared consts

I'm not convinced we'll actually want to use this, and there may be
another enable bit in SP_UNKNOWN_AB00, but it's nice to at least write
this down in case we want to try using it in the future.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5877>

freedreno/a6xx: Rename and document HLSQ_UPDATE_CNTL

It turns out that this clears CP_LOAD_STATE6 packets, including
disabling any pending loads for SS6_INDIRECT/SS6_BINDLESS (these loads
don't actually happen until the draw itself, and I'm not sure if they
happen if the state is unused by the shader) and marking constants and
UBO descriptors loaded with SS6_DIRECT as invalid. It's used very
differently from HLSQ_UPDATE_CNTL on a4xx from whence the name came, and
unlike on a4xx it's not readable, so this probably doesn't line up with
HLSQ_UPDATE_CNTL on a4xx.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5877>

clover: implements clEnqueueFillBuffer

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5897>

clover: add more cl_mem_object_type to pipe_texture_target mapping

It avoid unnecessary CL_INVALID_VALUE return from clGetSupportedImageFormats

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5897>

radv: add the custom border color BO to the list of buffers

The buffer was never added to the list of buffers. This might lead to
VM faults and GPU hangs.

Found this by luck.

Fixes: 57e796a12a8 ("radv: Implement VK_EXT_custom_border_color")
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/5884>

freedreno/registers: Rename SP_2D_SRC_FORMAT

This register contains information about the destination format, so
let's rename to SP_2D_DST_FORMAT.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Split clear and blit texture into different functions

Now that most of the state programming is in shareable helpers, we can
split emit_blit_or_clear_texture into emit_blit_texture and
fd6_clear_surface.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Don't take pipe_blit_info in emit_blit_dst

We only need a few fields and we'll want to use this in cases where we
don't have a pipe_blit_info.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Program RB_UNKNOWN_8C01 in setup helper

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Move CP_SET_MARKER to setup helper

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Move REG_A6XX_SP_2D_SRC_FORMAT programming to helper

Rename helper to emit_blit_setup().

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Program A6XX_SP_2D_SRC_FORMAT_COLOR_FORMAT based on dst format

It's a badly named register...

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Make blit_control helper a little more helpful

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Enable FMT6_10_10_10_2_UNORM blitting

Now that we correctly program the _DEST version for the blit
destination and use float16 internal format, these formats work with
the blitter.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Separate stencil sysmem clear fix

We need to clear with PIPE_FORMAT_S8_UINT.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Don't emit src state when clearing

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Consolidate computing blit_cntl

Compute the blit_cntl value in one place and group it with the
register writes.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Program SP_2D_SRC_FORMAT outside blit loop

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Set src and dst rects outside blit loop

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Don't set unknown bit when tiling differs

There is a bit here that's sometimes set, but it's generally not
related to whether tiling differs between src and dst.  Let's stop
setting it until we know more.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Split out src and dst setup helpers for blit

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Move fd6_ifmt into fd6_blitter.c

It's only used in this file.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

freedreno/a6xx: Don't blit with R2D_RAW

Map all formats to a valid ifmt.  FMT6_10_10_10_2_UNORM_DEST still
doesn't work on the blitter so keep that one on the u_blitter path.

Fixes
dEQP-GLES31.functional.draw_buffers_indexed.random.max_implementation_draw_buffers.*
with FD_MESA_DEBUG=nogmem.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5717>

turnip: drop GS clear path

We didn't know how to write layer id without GS, since that's the only way
to do it through VK/GL, and the blob didn't implement this clear case (and
failed cases where it was absolutely necessary). However now we know how to
set it after some educated guesses and looking at tess/geom traces, so the
GS path can be dropped.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5790>

turnip: clean up primitive output state

We only need to emit one set of primitive output registers. This may differ
from the blob, because it seems to try to allow using the same pipeline
with tess/geom enabled/disabled.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5790>

freedreno/regs: update primitive output related registers

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5790>

freedreno/ir3: Fix uninit var warning.

It's a decent bit of analysis to see that the initialization will always
happen, and my compiler isn't doing so in at least one configuration.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5834>

freedreno/ir3_cmdline: Fix an uninit var warning.

You could only access entry through the initialized path, but we can clean
up the compiler warning by not keeping the other var.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5834>

turnip: implement VK_EXT_private_data

Which is using base class's implementation.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5539>

turnip: Use the common base object type and struct.

v2. Define new helper function to avoid duplicated a pair of function calls.
v3. Move new helper functions to vk_object.h and call them.
v4. Merge 2 commits to use commomn base object type and struct into one.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5539>

tu: Fix wrong copies of sampler descriptor.

Found this with the following patch but it exists since adding ycbcr
sampler to the struct.

Fixes: d070a7ba0cfb11f1e01774b9dd3775ab7cd0c4ea
Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5539>

vulkan: Adds helpers for vk_object (de)alloation and (de)initialization.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5539>

gitlab-ci: reduce a630 runner load

They seem to be sometimes taking a while to boot, which is triggering CI
timeouts.  (Possibly tftp server in bad shape?)  Cut out non-essential
a630 CI jobs, and reduce the gles3/gles31 jobs to compensate.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5893>

meson/intel: add missing dep on git_sha1.h

Fixes: 805b32cab90547a576af ("intel: add identifier for debug purposes")
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5882>

zink: use type of src[0] for ntv store and load ops

in some cases (e.g., gl_ClipDistance) the nir_variable type doesn't match
the needed destination type, so we can simplify this code to just use
the destination type

fixes spec@glsl-1.10@execution@interpolation@interpolation-none-gl_backcolor-smooth-vertex

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5852>

zink: add lengthy comment and remove assert from discard_if ntv pass

as in the comment, while we may want to try verifying that discard will be
the last instruction in a block, it's a bit problematic given that other nir
passes we're doing may insert instructions after a discard as part of e.g.,
nir_opt_dead_cf in the process of removing another block

fixes shaders@glsl-fs-discard-04

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5852>

zink: handle ntv case of nested loop instructions more permissively

if the last instruction in a loop's body terminates a block, e.g., from
a nested loop with a jump as its final instruction, then no block will
have been started when returning to the original loop, and there's no need
to emit a branch

fixes shaders@glsl-vs-continue-in-switch-in-do-while

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5852>

zink: use right vulkan type for GL_PRIMITIVES_GENERATED queries

VK_QUERY_PIPELINE_STATISTIC_INPUT_ASSEMBLY_PRIMITIVES_BIT includes
primitives which won't get drawn due to e.g., not enough vertices emitted
by geometry shader

fixes spec@glsl-1.50@gs-emits-too-few-verts

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5533>

zink: only reset query pool on query end if current batch isn't in renderpass

reset can't be performed during a renderpass, so we need to defer that until a
time when we're definitely not in a renderpass, such as when we're starting a
new query or resuming a query

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5533>

zink: properly handle query pool overflows

inline a query result value to each query object so we can stash the partial
result just before we do a pool reset, which will always happen during the
suspend/resume query mechanism that swaps active queries from a flushed batch
to the next batch

once (or if) the "real" call to fetch query results is called, we can dump the
inlined value into the fetch value and return the full results

fixes mesa/mesa#3000

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5533>

zink: only stall during query destroy for xfb queries

xfb queries allocate vk buffer objects in the underlying driver which
can be deallocated while an xfb query is in-flight if we attempt to
defer it due to the way that gl xfb is translated to vk, so we need to
continue forcing this behavior in that case

for other query types, we can safely defer here until the current batch has
finished rather than block

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5533>

zink: use #define for number of queries per-pool

just to ensure we're consistent internally

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5533>

zink: rework query handling

this hooks up query objects to the batches that they're actively running on
(and the related fence) in order to manage the lifetimes of queries more
efficiently by calling vkCmdResetQueryPool only on init and when the query
pool has been completely used up. additionally, this resolves some vk spec
issues related to destroying pools with active queries

note that any time a query pool is completely used up, results are lost,
which is a very slight improvement on the previous abort() that was triggered
in that scenario

ref mesa/mesa#3000

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5533>

panfrost: Fix outmods on int to float conversions

No shader-db changes (Alyssa).

Signed-off-by: Italo Nicola <italonicola@collabora.com>
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5883>

docs/submittingpatches: add more than one `Cc: mesa-stable` example to the examples list

Starting with that very example :)

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5880>

docs/features: Update ASTC entries for Panfrost

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5856>

panfrost: Map PIPE_{DXT, RGTC, BPTC} to MALI_BCn

Mali (and Vulkan) uses D3D naming conventions for these formats where
Gallium/Mesa uses OpenGL names, but the formats are equivalent. sRGB is
communicated out-of-band on Mali; otherwise, it appears to be a 1:1
mapping.

On supported devices, this exposes GL_EXT_texture_compression_rgtc and
GL_ARB_texture_compression_bptc, so update features.txt

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5856>

panfrost: Filter compressed texture formats

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5856>

panfrost: Pipe in compressed texture feature mask

So we can query at run-time as part of Gallium's checks.

v2: More explicit naming.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5856>

panfrost: Add format codes for new compressed textures

Compressed formats line up with CONFIG_TEX_COMPRESSED_FORMAT_ENABLE
documented on
https://releases.linaro.org/archive/14.07/android/images/armv8-android-juno-lsk/

None of the new formats have been seen in the wild.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5856>

panfrost: Compact unused BO flag bits

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Suggested-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Remove panfrost_bo_access type

It's just whether it writes or not, which is already implied by the
presence/absence of a writer. So no need to track explicitly.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Remove PAN_BO_DONT_REUSE

Equivalent to SHARED.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Remove PAN_BO_COHERENT_LOCAL

Ancient relic from kbase. Panfrost kernel doesn't need this.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Merge PAN_BO_IMPORTED/PAN_BO_EXPORTED

Always checked together and really signal the same property from
different perspectives.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Index BOs from the BO map sparse array

Now we have a central store of them, so we may remove active_bo.

v2: Squash two patches together to prevent a race condition mid-series.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Add a sparse array to map GEM handles to BOs

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Fix write to free'd memory

No clue how this worked before.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Fixes: 82f18b713ac ("panfrost: Keep track of active BOs")
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

panfrost: Fix fence leak

When overwriting the writer, we need to release the old reference.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Fixes: 2dad9fde505 ("panfrost: Start tracking inter-batch dependencies")
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5859>

aco: add 32-bit integer addition to can_swap_operands

fossil-db (Navi):
Totals from 167 (0.12% of 135946) affected shaders:
CodeSize: 484892 -> 482628 (-0.47%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5695>

radv: use lower_shuffle_to_swizzle_amd

Affects a few shaders in Detroit: Become Human and Doom Eternal.

fossil-db (Navi):
Totals from 9 (0.01% of 135946) affected shaders:
CodeSize: 31188 -> 25096 (-19.53%)
Instrs: 6136 -> 4999 (-18.53%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5695>

nir/lower_subgroups: add lower_shuffle_to_swizzle_amd

masked_swizzle_amd can be much faster than shuffle.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5695>

nir/lower_subgroups: pass options struct to lower_shuffle

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5695>

aco: implement <32-bit masked_swizzle_amd

This is needed since we will be lowering some 8/16-bit shuffles to
masked_swizzle_amd.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5695>

aco: optimize some masked swizzles to DPP

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5695>

aco: read 0 from inactive lanes when using dpp

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5695>

panfrost: Enable framebuffer fetch

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Extend fetched framebuffer results

So NIR doesn't complain about invalid swizzles when reading a format
with less than 4 channels.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Always use SOFTWARE for pure formats

Otherwise we end up implicitly converting ints to floating point.
Likewise for floats which again have strange interactions.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Generate shader variants on framebuffer bind

If we keyed the shader for the framebuffer.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Use f2fmp for framebuffer lowering conversions

This allows the conversion to be removed when the output is needed as
f32 anyway, for example for highp framebuffer fetch.

v2: Also change operations such as i2i16 to i2imp (Alyssa).

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Stop keying on rt format when using native loads

Native loads are the same for any format, so we can use the same
shader variant for all framebuffer formats with a native load.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Implement texture_barrier

This is needed for KHR_blend_equation_advanced with a blend
barrier.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Use a 32-bit ld_color_buffer op when needed

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Set the z/s store intrinsic base correctly

When EXT_shader_framebuffer_fetch is used and only depth and/or
stencil are written, we can't rely on the first output being to
depth/stencil.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Support MRT in output load lowering

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Handle non-blend framebuffer lowering

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Emit a tilebuffer wait loop when needed

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Do the pan_lower_framebuffer pass later

The pass is useful for EXT_shader_framebuffer_fetch, not just blend
shaders, so we should do it with the other lowering passes in
midgard_compile_shader_nir.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Add a bitset of render targets read by shaders

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Add rt formats to shader state

load_output lowering will depend on the framebuffer formats.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Use the writeout tag for tilebuffer wait loops

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

pan/mdg: Handle tilebuffer wait loops

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Clamp pure int pixels

We need saturate, not wrap semantic. Could optimize to a .isat/.usat
modifier but that's for future.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

panfrost: Fix MALI_READS_TILEBUFFER

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>

nir: Add a base value to load_raw_output_pan

This is the render target the read instruction uses.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5755>