panvk: Don't expect subpasses to use all RTs

Some might use a subset, and that's fine, we just need to disable unused
RTs when beginning a subpass, which we do.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Fix an overflow on cmdbuf->state.clear

We assume the cmdbuf->state.clear array will have one entry per
attachment, but clearValueCount might be smaller if some attachments
are not cleared.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Handle input varyings without previous writes

Some input varyings might not be written by any of the active stages
preceding the stage reading the varying (e.g. gl_Layer should be set
to 0 when not written by vertex/geometry shaders). In this case, we can
insert a dummy varying attribute returning zero. This is actually what
the code intended to do, but 2 things were missing:

1. formats[NONE] is not mapping to the CONSTANT0 format
2. the offset and strides should always be set to 0 when using a
   CONSTANT0 attribute

All of this is needed to have the input attachments working. Indeed, we
use the nir_lower_input_attachments() pass which lowers input attachment
loads to texel fetches, and the txf operation is passed the layer_id
in its 3rd coordinate.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Avoid allocating sysvals UBOs when the pipeline has one

The static sysval UBO attached to the pipeline will be used anyway,
so let's not waste memory for sysvals if we can avoid it.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Fix multisample image copies

panvk_meta_copy_img2img() was assuming single-sample images, but the
spec actually allows copying multisample images, as long as the source
and destination agree on the number of samples.

Let's extend the img2img copy logic to support that case.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Fix panvk_pipeline_builder_upload_sysval()

Call panvk_sysval_upload_viewport_offset() instead of
panvk_sysval_upload_viewport_scale() in the PAN_SYSVAL_VIEWPORT_OFFSET
case.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Fix the pipeline binding logic

Right now, only one pipeline can be bound at any given time, instead of
one per bind-point. Fix the code so we can support compute operations.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Don't invalidate the vertex attributes when binding a new pipeline

This should be done when binding new vertex buffers.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Use the local batch variable when we have one

s/cmdbuf->state.batch->/batch->/ when the batch variable exists.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Make panvk_cmd_open_batch() return the new batch

So we can get replace the

   panvk_cmd_open_batch();
   batch = cmdbuf->state.batch;

sequence by

   batch = panvk_cmd_open_batch();

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Extend panvk_cmd_close_batch() to handle current_batch == NULL

This way callers don't have to check it.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Fix TLS initialization for multi-draw batches

If we queue multiple draws to the same batch, the TLS size should be
adjusted to cover the maximum TLS size requirement, not just the one
of the pipeline bound when the batch is closed. This requires keeping
TLS info attached to the batch and updating it every time a draw is
issued.

While at it, get rid of the wg_count state: we'll let vkCmdDispatch()
calculate the total WLS size.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Fix the static scissor/viewport case

Even if we can build a static viewport descriptor, we will need the
static scissor/viewport data to fill the viewport sysvals.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

panvk: Always allocate at least one BLEND descriptor for fragment shaders

Even there's no render target.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13077>

radv: remove vs_common_out:export_prim_id

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13062>

radv: remove unnecessary init of outinfo.export_prim_id for GS

When a geometry shader is present, not writing the gl_PrimitiveID is
undefined, so this is unnecessary. Note that this was never reached
on <= GFX8 because vs_common_out.export_prim_id from the GS key was
always FALSE.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13062>

radv: remove unnecessary radv_shader_info:{vs,tes}.export_prim_id

Use the VS outputinfo struct instead.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13062>

microsoft/clc: drop LLVM dependency to version < 12

Prior to LLVM 12, SmallVector requires 2 template arguments.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: c4755a7c32e66c ("microsoft/clc: Support SPIR intermediates in the compilation APIs")
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13083>

radv: remove unnecessary vs_common_out.export_layer_id

Same as the viewport index, the driver will emit 0 for the PS input
in this case.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13061>

radv: remove unnecessary vs_common_out.export_viewport_index

This was only used when the FS needs the viewport index if it's not
exported by the previous stage. Though, this is actually useless
because the driver will emit 0 for the PS input in this case.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13061>

radeonsi/test: add sanity checks

Verify that DISPLAY is there and deqp-runner is at least 0.9.

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13021>

radeonsi/test: add Raven expected results

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13021>

radeonsi/test: add --gpu to select the GPU to test

And specify DRI_PRIME + WAFFLE_GBM_DEVICE to avoid testing the wrong GPU.

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13021>

radeonsi/test: sanitize output_folder

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13021>

radeonsi/test: allow to specify a baseline folder

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13021>

radeonsi/test: print default values in help

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13021>

radv: don't require a GS copy shader to use the cache with NGG VS+GS

Fixes dEQP-VK.pipeline.cache.no_cache_control.readwrite_cache.pipeline_from_get_data.vertex_stage_geometry_stage_fragment_stage

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Fixes: 7e664a53835 ("radv: Don't generate GS copy shader when the pipeline has NGG.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12741>

radv: remove unused radv_nir_compiler_options fields

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13067>

radv: cleanup uses of VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT

Use the pipeline key instead of this mess.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13067>

util/cache: test simple cache put and get between instances

To make sure we are not just using the in-memory cache index for
the single file cache, we test adding and retriving cache items
between two different cache instances.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12925>

v3d: add R10G10B10X2_UNORM to format table

Fixes following piglit fails:
spec@ext_framebuffer_object@fbo-blending-formats
spec@ext_framebuffer_object@fbo-blending-formats@GL_RGB10

Cc: mesa-stable
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13051>

lavapipe: Fix crashes with transform feedback when using VK_WHOLE_SIZE

llvmpipe expects valid size parameter, and when just VK_WHOLE_SIZE is
passed very bad things can happen.
This was handled specially before, but got dropped when lavapipe was
converted to use the generated command queue.

Fixes: eb7eccc76f0a ("lavapipe: Use generated command queue code")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13036>

anv: assert that legacy_scanout isn't used with explicit modifiers

these should be mutually exclusive

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12919>

iris: Document push constants allocation

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13014>

intel: Add and use max_constant_urb_size_kb

This knowledge was repeated in multiple places so move the values to
intel_device_info struct.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13014>

spirv: don't bother initializing variables to Undef

If an OpVariable's initializer is undef, there is no need to
initialize the variable.

v2: Comment the code (Caio)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13030>

spirv: workaround LLVM-SPIRV Undef variable initializers

The LLVM-SPIRV translator creates variables with initializers, but
most of those are actually undef initializers. We can just skip
composites that are entirely made of undefs, but for partially undefs,
we will still zero initialize.

v2: Rename wa_llvm_spirv_undef_initializer to wa_llvm_spirv_ignore_workgroup_initializer (Caio)
    Limit workaround to OpenCL (Caio)
    Make workaround clearer (Caio)

v3: Only apply workaround on workgroup storage (Caio)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13030>

spirv: avoid shadowing local variable

v2: rename s/eval/elem_val/ (Caio)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13030>

clover: Rename module -> binary, because C++20 makes module a keyword

Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12273>

clover: Delete unused 'e' exception reference vars

Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12273>

clover: std::result_of is deprecated in c++17 and removed in c++20

Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12273>

i915g: Use the non-vbuf code path by default to fix index overflows.

We were assertion failing on some large draws due to indices >16bits,
despite asking draw to limit the max indices.  I haven't managed to track
it down, so flip us back to the older, non-index drawing path that doesn't
hit this bug until it can get fixed.  Leave an I915_DEBUG=vbuf flag around
so we can look into this later.

This is a pretty big performance hit for vertex shaders.  Using glmark2 -b
build:use-vbo=true:

i915g-vbuf:     211 fps
i915g-nonvbuf:  185 fps
i915c:          41 fps

Given how massively better i915g still is than i915c (llvmpipe VS instead
of the classic swrast interpreter), I think it's still worth it to get
i915g correct before we fix this perf regression.

Fixes: #4971
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13052>

i915g: Unifdef VBUF_MAP_BUFFER.

You do want to stream the vertices out to the WC mapping, as the code has
been doing, rather than writing into malloc and doing a memcpy later and
wasting cache space.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13052>

i915g: Remove dead VBUF_USE_POOL code.

Not defined anywhere, and the members it's setting up don't exist.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13052>

i915g: Improve debug output for the fresh-batch overflow case.

Much more useful info for dEQP-GLES2.functional.buffer.write.random.0 than
"i915_vbuf_render_draw_elements: Assertion `0' failed."

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13052>

ci/vc4: add piglit timeout

Acked-by: Emma Anholt <emma@anholt.net>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13065>

turnip: Disable VK_EXT_display_control.

The common code fails dEQP-VK.wsi.display_control.register_device_event
due to having a stub NOT_IMPLEMENTED return, and thus fails the CTS.  This
is one of our last failures, so disable the extension until it can get
finished off, so we can unblock passing the CTS.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13010>

aco: try forming clauses even if reg_pressure exceeds

This patch allows to form clauses even if the register pressure
is at the limit with the effect that VMEM instructions are less
scattered after the first clause in a Block.
It respects the previous clause size to avoid excessive moving
of VMEM instructions.
VMEM_CLAUSE_MAX_GRAB_DIST is further reduced to compensate
some of the effects.

Totals from 28922 (19.26% of 150170) affected shaders: (GFX10.3)
VGPRs: 1546568 -> 1523072 (-1.52%); split: -1.52%, +0.00%
CodeSize: 117524892 -> 117510288 (-0.01%); split: -0.08%, +0.07%
MaxWaves: 605554 -> 611120 (+0.92%)
Instrs: 22292568 -> 22291927 (-0.00%); split: -0.10%, +0.09%
Latency: 488975399 -> 490230904 (+0.26%); split: -0.06%, +0.32%
InvThroughput: 117842300 -> 116521653 (-1.12%); split: -1.15%, +0.03%
VClause: 541550 -> 522464 (-3.52%); split: -9.73%, +6.20%
SClause: 718185 -> 718298 (+0.02%); split: -0.00%, +0.02%
Copies: 1420603 -> 1386949 (-2.37%); split: -2.64%, +0.27%
Branches: 559559 -> 559278 (-0.05%); split: -0.06%, +0.01%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10896>

aco: make clause-forming depend on the number of moved instructions

This allows more aggressive clause-forming in presence of
larger def-use distances. To compensate for the effect,
VMEM_CLAUSE_MAX_GRAB_DIST was decreased.

Totals from 5788 (3.85% of 150170) affected shaders: (GFX10.3)
VGPRs: 483960 -> 475272 (-1.80%); split: -1.82%, +0.02%
CodeSize: 59661240 -> 59669084 (+0.01%); split: -0.01%, +0.02%
MaxWaves: 70408 -> 71450 (+1.48%); split: +1.51%, -0.03%
Instrs: 11222417 -> 11224479 (+0.02%); split: -0.01%, +0.03%
Latency: 349397104 -> 349298602 (-0.03%); split: -0.03%, +0.00%
InvThroughput: 88584832 -> 87762262 (-0.93%); split: -0.93%, +0.00%
VClause: 168905 -> 177089 (+4.85%); split: -0.48%, +5.32%
SClause: 375795 -> 375767 (-0.01%); split: -0.01%, +0.01%
Copies: 840298 -> 840231 (-0.01%); split: -0.04%, +0.03%
Branches: 373265 -> 373278 (+0.00%); split: -0.00%, +0.00%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10896>

aco: stop scheduling if clause-forming fails

This avoids unintended reordering of VMEM instructions.
It is also highly unlikely that we find more independent
instructions before previous clause-related instructions.

Totals from 1921 (1.28% of 150170) affected shaders: (GFX10.3)
VGPRs: 103832 -> 103736 (-0.09%); split: -0.10%, +0.01%
CodeSize: 8695560 -> 8706000 (+0.12%); split: -0.03%, +0.15%
Instrs: 1643752 -> 1646349 (+0.16%); split: -0.04%, +0.20%
Latency: 26755527 -> 26614645 (-0.53%); split: -0.67%, +0.14%
InvThroughput: 7226604 -> 7204809 (-0.30%); split: -0.39%, +0.08%
VClause: 46536 -> 46201 (-0.72%); split: -0.81%, +0.09%
SClause: 47910 -> 47769 (-0.29%); split: -0.43%, +0.14%
Copies: 94647 -> 94558 (-0.09%); split: -0.26%, +0.17%
Branches: 36843 -> 36847 (+0.01%); split: -0.00%, +0.01%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10896>

glx: Don't strip off window/pixmap support from float fbconfigs

The X server doesn't get this wrong. It's not the client's job to
correct what the server says here. And if anyone ever implements HDR for
X11, you might in fact want to be able to use floats with a window.

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13002>

glx: Clarify a debug message

In GLX a "tag" usually means a context tag, "fbconfig attribute" is a
bit more obvious.

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13002>

glx: Move __glFreeAttributeState next to its one caller

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13002>

glx: Remove some unused declarations from glxclient.h

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13002>

aco/ra: don't copy linear VGPRs within CF in get_reg_create_vector()

Fixes: 6ed18749de52d1f24b23fad266eb3e8b46702752 ('aco: allow live-range splits of linear vgprs in top-level blocks')
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13058>

radv: pass the pipeline key to the backend compilers

It exactly matches the shader keys now. Everything was copied from
the pipeline key to the shader keys.

There is still some work to completely remove radv_shader_variant_key.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13032>

radv: rework layout of radv_pipeline_key

To match radv_shader_variant_key.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13032>

radv: store the CS subgroup size to radv_shader_info

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13032>

radv: store the topology instead of the output primitive type in the key

To match the pipeline key.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13032>

panfrost/ci: Test panvk on Mali G52

Just run some selected tests for now because we miss a lot of
functionality, which would cause so many crashes that the runs
aren't practical.

Once the core functionality is implemented, we can switch to the master
case list with skips.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13016>

panfrost/ci: Move CI files to src/panfrost

In preparation for testing panvk.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Acked-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13016>

panvk: Implement vkCmdBlitImage()

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12961>

pan/blit: Fix 3D blittering

Fixes several problems in the pan_blit() logic:

1. We actually need the reciprocal of the depth scaling in z_scale (maybe
   we should rename this field z_scale_rcp to make it clear)

2. When Z end < Z start we should remove one to the cur_layer/layer_offset
   instead of doing it on the last_layer field, otherwise there's an
   off-by-one error

3. The Z src offset should be adjusted to account for scaling. If we don't
   do that we won't sample from the right layer when upscaling.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12961>

pan/blit: Let the caller offset the start/end coords passed to the blitter

Since we have no guarantee that start < end, we can't really tell to
which one the offset applies to. Let the caller take care of that.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Acked-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12961>

lima: Fix glFrontFace handling

Bit 12 of render->aux1 is GL_CCW/GL_CW. For GL_CCW (default of glFrontFace) we have
to set that bit active.

This is not what the blob does and what the original reverse engineering documentation
says. The blob sets this value inverted and does some bogus negation of the fragment
shaders gl_FrontFacing variable instead.

Anyway, doing it this way does not cause regressions but fixes
dEQP-GLES2.functional.shaders.builtin_variable.frontfacing and 4 piglit tests.

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7690>

lavapipe: move to new shared features/properties

Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12953>

lavapipe: enable vulkan 1.2 support.

The remaining extensions are optional features, just turn on vk 1.2
with them reporting as off.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12953>

gallivm/nir: fix subgroup invocation read.

Again if you get passed an invoc but the exec mask has the
active lane somewhere other than at 0, then if we have an
invoc we should find the active lane and extract the value
from invoc rather than using the idx.

This fixes a bunch of VK 1.2 subgroup tests once 1.2 is enabled:
dEQP-VK.subgroups.ballot_broadcast.compute.subgroupbroadcast_nonconst*

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12953>

lavapipe: move to 1.2 features/properties structs.

This just adds all the wrappers in the right places hopefully

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12953>

lavapipe: add support for KHR_shader_float_controls

These pass all the CTS tests, though not sure how useful they are.

[airlied: these may need some work in the future depending on app expectations]

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12953>

features: mark off bindless texture for zink

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: export PIPE_CAP_BINDLESS_TEXTURE

this is the 6th descriptor set being bound, so don't even advertise it
if 6 sets can't be bound

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: implement bindless textures

this works by tracking 1024-member arrays of images and textures using idalloc
for indexing. each idalloc id is an index into the array that is returned as a handle,
and this handle is used to index into the array in shaders.

in the driver, VK_EXT_descriptor_indexing features are used to enable updates on the live
bindless descriptor set and leave unused members of the arrays unbound, which works as
long as no member is updated while it is in use. to avoid this, idalloc ids must cycle through
a batch once the image/texture handle is destroyed before being returned to the available pool

in shaders, bindless ops come in one of two types:
- i/o variables
- bindless instructions

for i/o, the image/texture variables have to be rewritten back to the integer
handles which represent them so that the successive shader stage utilizing them
can perform the indexing

for instructions, the src representing the image/texture has to be rewritten as a deref
into the bindless image/texture array

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: hook up VK_EXT_descriptor_indexing

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: handle bindless images and samplers in ntv

these are going to come through as direct variable derefs, so it's simple
to handle the functionality by reusing the same codepath to generate image
types

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: refactor some shader image code to make it reusable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: don't add resource to pending barrier set if no barrier will be generated

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: split out bvci creation from object creation

make this reusable like the others

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12855>

zink: force lazy descriptor set rebinds if pipeline compatibility changes

if the pipeline layout changes, it's technically illegal to not rebind the descriptor
set even if it hasn't changed

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12931>

zink: remove duplicated struct member set

this should only happen in zink_descriptors_update_lazy_masked

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12931>

ac/surface: correct a comment about DCC image stores

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13013>

ac/surface: use DCC compatible with image stores for < 4K resolutions

We don't have to use the special DCC settings for lower resolutions.

This will cause corruption if X and an windowed app use different Mesa
versions. The fix is to restart the X server. I expect to get false bug
reports due to this.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13013>

radeonsi: remove the unused cs parameter from remaining packet functions

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13015>

radeonsi: remove the unused cs parameter from radeon_set_uconfig_reg

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13015>

radeonsi: remove the unused cs parameter from radeon_set_sh_reg

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13015>

radeonsi: remove the unused cs parameter from radeon_set_(config|context)_reg

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13015>

radeonsi: remove the unused cs parameter from radeon_emit_array

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13015>

radeonsi: remove the unused cs parameter from radeon_emit

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13015>

docs: mark off GL_EXT_multisampled_render_to_texture for zink

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: implement GL_EXT_multisampled_render_to_texture

this requires allocating and using a lazily-allocated msaa surface as a transient
attachment for the base render operation, resolving it

...except vulkan has no "replicate" renderpass attachment mechanism, so for now we're
just gonna allocate a transient surface and hang on to it. sorry tilers!

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: stop setting nr_samples for shader image surface creation

that's not what this should be used for

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: move fb attachment init to new function

this is going to get more complicated

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: remove fbfetch layout thingy from zs renderpass init

zs doesn't do fbfetch

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: reorder draw state updates

starting a renderpass can trigger recursion and other draws, so do all
state updates/binds after the renderpass has begun

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: improve handling of buffer rebinds using tc info

tc rebind info can't distinguish between buffers bound multiple times
in a given bind point (e.g., for multiple vertex buffers), so ensure all
rebinds are processed and the global rebind isn't triggered if multiple rebinds
are processed for a given buffer

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: add ZINK_BIND_TRANSIENT

this is used to create lazily allocated images for use as transient
renderpass attachments

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: add ZINK_HEAP_DEVICE_LOCAL_LAZY

this is for lazily allocated resources

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

zink: repack zink_render_pass_state

saves a couple wasted bits

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12934>

spirv: don't fail on CapabilitySubgroupDispatch if supported

Since only Anv uses the value, I'm only enabling this on anv.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 518693c3eca195 ("spirv: Handle the SubgroupSize execution mode")
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13034>

zink: split vk debug logging into separate functions

being able to breakpoint these easily is crucial to debugging, and it sucks to
always have to go back and check the line numbers, so now there's easily-rememberable
functions for it

Reviewed-by: Hoe Hao Cheng <haochengho12907@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13042>