zink: add a bits union for zink_render_pass_state comparisons

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: reuse rt layout functions to refactor zink_render_pass_attachment_get_barrier_info()

maybe slightly easier to read?

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: break out color rt layout calc

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: break out zs renderpass layout calc

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: break out renderpass attachment load op calc

no functional changes

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: split out rp_changed to be more granular for dynamic render

sometimes a state change MAY require a renderpass change, but this change
will not require splitting the current renderpass

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: add an assert to verify that deferred clears aren't added in renderpass

this would be very, very broken

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: remove redundant renderpass update check for clears

this should be handled already in the clear code

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: unflag ctx->rp_changed after dynamic render update

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: do dynamic render implicit clear updates on rp_changed

this should be flagged correctly now

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: flag renderpass change if renderpass clears change

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: don't flag renderpass change on depth write change as often

this is only "sometimes" required and only if dynamic render isn't being
used

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

zink: don't try to start a renderpass before CmdClearAttachments

it should be impossible to reach this point if a renderpass is not active

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17640>

llvmpipe: ignore multisample state for alpha_to_one

this is more accurate and fixes usage with lavapipe

cc: mesa-stable

Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17627>

vulkan/wsi: fix multiple acquires for sw without mit-shm

in this case, lying about having multiple images and then returning the
same image every time doesn't work, so use the busy flag
and return an available image when possible

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17590>

radv, aco: Don't lower 16bit isign.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17440>

radv: Remove ineg from lower_bit_size_callback callback.

It's always lowered to isub.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17440>

radv,aco: Don't lower and vectorize 16bit iabs.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17440>

aco: remove unnecessary exp_cnt increments

update_counters() already does this.

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/17257>

aco: Initialize spill_ctx members in constructor.

Fix defects reported by Coverity Scan.

Uninitialized scalar field (UNINIT_CTOR)
uninit_member: Non-static class member sgpr_spill_slots is not
initialized in this constructor nor in any functions that it calls.
uninit_member: Non-static class member vgpr_spill_slots is not
initialized in this constructor nor in any functions that it calls.

Fixes: 7d340449086 ("aco: refactor VGPR spill/reload lowering")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17583>

zink: use gfx_barrier for synchronization when binding buffer descriptors

this will preserve existing shader stages in the access mask and avoid desync
later on

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17639>

zink: fix viewport count tracking

the number of viewports in use is based on the outputs of the last vertex
stage, not the viewports passed by the state tracker

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17639>

zink: call bind_last_vertex_stage() when binding vs

ensure all the necessary updates are flagged

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17639>

zink: improve oom error message

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17639>

zink: fix crashing on bo allocation failure

list init occurs after mem allocation, so check this first

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17639>

zink: handle null samplerview in get_imageview_for_binding()

this is legal

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17639>

zink: handle max_vertices=0 in geometry shader

this is a weird corner case where glsl permits a zero value, so clamp to 1
and then don't emit any vertices to avoid driver hangs

affects:
dEQP-GL45-ES31.functional.geometry_shading.emit.points_emit_0_end_0

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17639>

v3d/v3dv: use NIR_PASS(_

Instead of NIR_PASS_V, when possible.

This was done recently on anv (see commit ce60195ec and MR#17014)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

broadcom/compiler: make several passes to return a progress

Two advantages:
   * When using NIR_DEBUG=nir_print_xx, will print outcome only if
     there is a change
   * We can use NIR_PASS(_, ...) instead of NIR_PASS_V, that has
     slightly more validation checks.

This includes:
  * v3d_nir_lower_image_load_store
  * v3d_nir_lower_io
  * v3d_nir_lower_line_smooth
  * v3d_nir_lower_load_store_bitsize
  * v3d_nir_lower_robust_buffer_access
  * v3d_nir_lower_scratch
  * v3d_nir_lower_txf_ms

As we are here we also simplify some of them by using the
nir_shader_instructions_pass helper.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

broadcom/compiler: removed unused function

It is not even implemented.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

v3dv/pipeline: use nir_shader_instructions_pass when lowering layout

The trigger for this commit was when we found that we were not calling
nir_metadata_preserve when lowering the layout code. But then I found
that it would be better to just update the code to use
nir_shader_instructions_pass, so we can avoid to manually:
   * Initialize the nir_builder
   * Call nir_foreach functions (we pass the callback)
   * Call nir_metadata_preserve functions (that as mentioned we were not calling)

We also get a nice cleanup of several functions by reducing the number
of parameters (we pass a state struct).

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

nir: call nir_metadata_preserve at nir_remove_unused_io_vars

Without it we got a metadata assert:
deqp-vk: ../src/compiler/nir/nir_metadata.c:108: nir_metadata_check_validation_flag: Assertion `!(function->impl->valid_metadata & nir_metadata_not_properly_reset)' failed

if we try to use NIR_PASS(_, instead of NIR_PASS_V (that among other
things, do more validations).

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

broadcom/compiler: use NIR_PASS for nir_lower_vars_to_ssa at v3d_optimize_nir

There's no reason to not take into account progress at that point.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

broadcom/compiler: wrap nir_convert_to_lcssa with NIR_PASS_V

So we get it included with the NIR_DEBUG=print_xx debug options.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

broadcom: document which debug options are specific for a given driver

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

broadcom: add no_merge_jobs debug option

To avoid having a custom getenv on v3dv.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

v3dv/device: add comment on depthClamp feature

As we can't support it, and I need to recheck now and then when I read
that code.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>

vulkan: allow null descriptor set layouts in pipeline layouts

VK_NULL_HANDLE descriptor set layouts are allowed when creating pipeline
layouts without VK_PIPELINE_LAYOUT_CREATE_INDEPENDENT_SETS_BIT_EXT.

From VUID-VkPipelineLayoutCreateInfo-graphicsPipelineLibrary-06753:

> If graphicsPipelineLibrary is not enabled, elements of pSetLayouts
> must be valid VkDescriptorSetLayout objects

From VUID-VkPipelineLayoutCreateInfo-pSetLayouts-parameter:

> If setLayoutCount is not 0, pSetLayouts must be a valid pointer to an
> array of setLayoutCount valid or VK_NULL_HANDLE VkDescriptorSetLayout
> handles

Signed-off-by: Ricardo Garcia <rgarcia@igalia.com>
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17629>

v3dv: fix offset reported by vkGetImageSubresourceLayout

This represents an offset from the actual start of the image data,
not from the start of the memory allocation bound to the image.

Fixes:
dEQP-VK.image.subresource_layout.*

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17648>

v3dv: remove assert that poolSizeCount must be larger than 0

From the Vulkan spec:

   "If poolSizeCount is not 0, pPoolSizes must be a valid pointer to an
    array of poolSizeCount valid VkDescriptorPoolSize structures"

So 0 is actually allowed and there is a CTS to check it is handled gracefully.

Fixes:
dEQP-VK.api.descriptor_pool.zero_pool_size_count

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17648>

v3dv: handle VK_STRUCTURE_TYPE_DRM_FORMAT_MODIFIER_PROPERTIES_LIST_2_EXT

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17616>

nir: convert unused mesh outputs to shared memory

Otherwise reads from output in one subgroup may not see
writes from other subgroups. Temp variables are later converted
to scratch, so even within one subgroup we may not see correct values.

Test case in https://gitlab.freedesktop.org/mesa/crucible/-/merge_requests/115

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17517>

ci: wine-apitrace bump to 11.1

In the future, bump will be needed to update non-VK tags as wine traces get
used elsewhere.

Reviewed-by: Andres Gomez <agomez@igalia.com>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17623>

v3d/ci: (partially) revert expected list

This reverts the traces expected list changed in f3579a62e9b
("v3d/v3dv/ci: update expected results").

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17647>

zink: default to lazy/templated descriptors

cached mode was great 2 years ago when template support was less widespread,
but now that templates are everywhere, caching is less performant in
every scenario

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17636>

zink: add an explicit ZINK_DESCRIPTOR_MODE_CACHED

no functional changes, but now AUTO is no longer a mode that gets used

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17636>

zink: delete ZINK_DESCRIPTOR_MODE_NOFALLBACK

this was useful for testing but is going to be less useful soon

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17636>

zink: move descriptor_mode to be a global

this can't change once it's been set, so having it per-object is pointless

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17636>

nir/loop_unroll: clean up after complex_unroll_single_terminator()

Previously we would just unroll the loop one extra iteration and let
other optimisation passes clean up the mess. This worked to a degree
but if the loop happened to be nested inside another loop we would
end up with phi chains that would block other passes from being able
to do the cleanup.

With this commit we explicitly clone the variables create by lcsaa
and insert them directly in the last continue branch after we are done
unrolling. With this optimisation passes can recognise both sides
of the if output the same values and can progress further.

Help with the issues described in:
https://gitlab.freedesktop.org/mesa/mesa/-/issues/6051

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17611>

lavapipe: state latest conformance tests passed

Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Cc: mesa-stable "22.1"
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17643>

turnip: Disable LRZ fast-clear for gen1 and gen2

LRZ fast-clear works on all gens, however blob disables it on
gen1 and gen2. We also elect to disable fast-clear on these gens
because for close to none gains it adds complexity and seem to work
a bit differently from gen3+. Which creates at least one edge case:
if first draw which uses LRZ fast-clear doesn't lock LRZ direction
the fast-clear value is undefined.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6829

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17599>

ci/freedreno: Switch a630 to manual/disabled for lab maintenance.

We're getting several more 630s in the lab, but need a bit of time to swap
out some broken old ones and stabilize the new ones.  Fritz thinks this
should be done in an hour or so, but I want to turn off the CI for main so
that we don't block anyone else.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17641>

venus: reenable VK_EXT_extended_dynamic_state2

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17582>

venus: check dynamic state for raster enablement

We should not scrub raster dedicated states when dynamic state includes
VK_DYNAMIC_STATE_RASTERIZER_DISCARD_ENABLE.

Test:
- dEQP-VK.pipeline.extended_dynamic_state.*_raster
- dEQP-VK.api.pipeline.pipeline_invalid_pointers_unused_structs.graphics

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17582>

crocus: fail query begin if upload allocation fails.

This is the only place I think I can see this crashing.

Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17615>

Zink: Fix clear being missed when using emulated draws in zink_blit

zink_kopper_acquire_readback will flush any outstanding clears, this means that
the current clears need to be applied first before calling zink_kopper_acquire_readback.
This was already done for native_blit and native_resolve, also do this for the emulated draw path.
Seen as intermittent failures in cts case GTF-GL33.gtf21.GL2FixedTests.buffer_clear.buffer_clear.

Cc: mesa-stable
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17631>

zink: Fix spirv stream 0 vertex emit for multistream shaders

Spirv spec does not allow the use of OpEmitVertex or OpEndPrimitive when there are multiple streams.
Instead emit the multi-stream version of these with stream set to 0.
This issue was seen when testing cts case KHR-GL46.transform_feedback.draw_xfb_stream_test

Fixes: 35e346f4280 ("zink: handle vertex streams")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17513>

anv: Do depth/stencil optimization for dynamic depth/stencil

When we made depth/stencil dynamic, we lost the optimization.  This is
particularly important for cases where the stencil test is enabled but
never writes anything as certain combinations with discard can cause
the stencil write (which doesn't do anything) to get moved late which
can be a measurable perf hit.  According to 028e1137e67b ("anv/pipeline:
Be smarter about depth/stencil state", it was a couple percent for DOTA2
on Broadwell back in the day.  No idea how it affects current titles.

This may also improve the depth/stncil PMA workarounds on Gen8 and Gen9
since they're now looking at optimized depth/stencil state.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

vulkan/graphics_state: Improve the depth/stencil optimization code

We now disable stencil writes when the stencil test is disabled or when
both stencil write masks are zero.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: Switch to using common dynamic state tracking

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: Stop merging DEPTH_STENCIL state

Now that we've stopped trying to do dynamic stuff up-front, we're only
merging in one bit: DoubleSidedStencilEnable.  There's no point in all
the merging code for one bit which is a constant anyway.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: Rework setting primitive topology

For one thing, we were deceptively setting it wrong in genX_cmd_buffer.c
and then overwriting it in each of of gfx7_cmd_buffer.c and
gfx8_cmd_buffer.c.  Pull it all into genX_cmd_buffer.c so it's no longer
duplicated.  Also, stop doing the PATCHLIST conversion in anv_pipeline.c
and just store the number of patch vertices.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: Convert to using vk_graphics_pipeline_state

This attempts to be a pretty minimal refactor.  We just switch all the
state emit code to use the new graphics pipeline state object.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: remove pipeline setting dynamic state

Following our rule that anything that can be dynamic should be set
(and is this case is already) in the following files :

   - genX_cmd_buffer.c
   - gfx8_cmd_buffer.c
   - gfx7_cmd_buffer.c

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: Stop recording sample locations per-sample-count

The only reason why we recorded them per-sample-count is because Intel
hardware is weird starting with Broadwell.  The API, requires that the
dynamic sample pattern be reset every time the sample count changes so
we only need to record the pattern for the current sample count.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: Handle vertex buffer sizes in anv_CmdBindVertexBuffers2

There's no good reason to defer figuring out the size until we emit the
packet.  We know everything when the bind happens.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

anv: Add an instance multiplier to anv_pipeline

This lets us get rid of the view_mask and pipeline replication logic
repeated all over everywhere.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17564>

spirv: handle phis decorated with RelaxedPrecision

If the driver can do 16-bit ALU ops, then store RelaxedPrecision phi
values into 16-bit NIR variables with downconverts/upconverts on the way
in/out.

This has no impact on shader-db on freedreno (not that we have a ton of
GLES content there), but it does cause an ANGLE-translated CTS shader on
vulkan to get consistent conversions between two copies of a value, and
avoid a test bug.

Reviewed-by: Emma Anholt <emma@anholt.net>
Closes: #6585
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14018>

ci/bare-metal: Add timeouts to the shell commands called in fastboot.

It seems that we sometimes stall out executing "fastboot boot", and if
that happens we want to reboot the board and try again.

Fixes: #6682
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17607>

radv: don't update misaligned_mask in CmdBindVertexBuffers2

This can't do it in the loop because it doesn't easily know what
attributes use a binding.

We could do it in a separate loop, but there's no point, especially since
zink does CmdSetVertexInputEXT() after CmdBindVertexBuffers2().

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Fixes: c335a4d70ed ("radv: dynamically calculate misaligned_mask for dynamic vertex input")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17521>

radv: fix vbo_bound_mask indexing

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6812
Fixes: 1b8bdecf6e0 ("radv: add a mask of bound descriptor buffers for dynamic vertex input")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17521>

r300: disable PIPE_CAP_PREFER_IMM_ARRAYS_AS_CONSTBUF

We can't do uniform buffer objects and from the hardware
perspective constants (uniforms) and immediates are treated in
the same way. They are uploaded together and fit together into the
(rather low) total constant limit. Therefore, there is actually no
advantage in converting immediates to uniforms, and a whole lot of
disadvantages (less possible optimizations and no inlining).

Fixes the dEQP regressions from https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16770

The tests failed because when there is an indirect array access the
compiler inserts a big if ladder and converting the temp to a uniform
means more instructions because the ifs cant be lowered at least
partially to selects. It is particularly visible, because the code
NIR currently emits for the indirect access doesn't really fit the
hardware, see: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6366

No change in my shader-db, so this concerns the tests only.

Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17576>

vulkan: Detect pNext chain loops in vk_foreach_struct() (v2)

This implements the "tortoise and the hare" algorithm for detecting
cycles in graphs.  We use the caller's iterator as the hare and our own
internal copy as the tortoise.  Conveniently, VkBaseOutStructure (and
VkBaseInStructure which is identical except the pointer type on pNext)
have a pointer we can use for the tortoise and an sType which we can use
for a counter to ensure we only increment the tortose every other loop
iteration.

There are more efficient algorithms than tortoise and hare but they
require allocating memory for something like a hash set of seen nodes.
Since this for debug purposes only, it's ok for it to be a bit
inefficient in the case where it hits the assert.  In the usual case of
no loops, it's the same runtime efficiency as the unchecked version
except that it does a tiny bit of math and 50% more pointer chases.

Version 1 worked fine with clang and with GCC 12.1 with -O0 but not with
optimizations. See https://gitlab.freedesktop.org/mesa/mesa/-/issues/6895.
Unfortunately, the first version required modifying a temporary declared
const inside the for loop and that seems to have been the problem.  This
version instead has an iterator struct which is managed by an outer for
loop and the inner for loop exists to declare the user's requested
iteration variable and manage the actual iteration.  Because the outer
for loop is effectively `for (bool done = false; !done; done = true)`,
it will execute exactly once, regardless of the inner loop, so break and
continue inside the inner loop should work the same as if it's a single
for loop.

The other major difference with the new version is that the code is the
same for debug and release except the half_iter and loop check are gone.
I've verified by hand that this produces virtually identical code to the
old simple iterators on both GCC andl clang with an optimized build.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17630>

Use vk_foreach_struct_const where needed

We're about to make it so that the compiler warns/errors if you use the
wrong iterator macro.  Fix up a bunch of places where someone used the
wrong one before we break anything.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17630>

vulkan: Copy the depth/stencil state optimization code from ANV

Instead of having stencil writes as an out parameter of the optimization
function, we add a new write_enable field for stencil that's equivalent
to the similarly named field for depth.  This doesn't mean drivers must
actually support disabling stencil writes independently but the
information may be helpful on some hardware.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

vulkan: Add a common implementation of vkCmdSet*

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

vulkan: Add a common data structure for dynamic states

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

util: Constify __bitset_test_range

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

vulkan: Add data structures to store all graphics state

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

vulkan: Add standard sample locations

This looks a bit funny because we define the struct in the .c file but
that will be fixed in the next commit.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

vulkan: Add an enum for all dynamic graphics states

Also, add a helper function to turn VkPipelineDynamicStateCreateInfo
into a bitset.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

vulkan: Add a vk_limits.h file for runtime limits

Individual driver limits may be smaller than these.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17328>

aco: requires Exact for p_jump_to_epilog

Otherwise, in presence of p_exit_early_if the main FS will always
jump to the PS epilog regardless the exact mask.

This fixes dEQP-VK.draw.renderpass.shader_invocation.helper_invocation
and few vkd3d-proton regressions when PS epilogs are forced.

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/17617>

v3d/v3dv/ci: update expected results

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17632>

aco: fix assertion in insert_exec_mask

The exec mask might also be of type mask_type_loop.

Fixes: d068eb53e84ca1e44ad96c31dab63476880b3c72 ('aco/insert_exec_mask: optimize top-level transition to exact before demote')
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17402>

aco: Avoid live-range splits in Exact mode

Because the data register of atomic VMEM instructions
is shared between src and dst, it might be necessary
to create live-range splits during RA.
Make the live-range splits explicit in WQM mode.

Totals from 7 (0.01% of 134913) affected shaders: (GFX10.3)
Latency: 17209 -> 17210 (+0.01%)

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15347>

aco: avoid unnecessary copies in emit_wqm()

No fossil-db changes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15347>

turnip: expose VK_EXT_shader_module_identifier

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17614>

turnip: Remove an unnecessary assert.

The assertion is already in the common implementation.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17614>

ac/llvm: Remove load_vertex_id handling

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17539>

aco: Remove dead nir_intrinsic_load_vertex_id case

This intrinsic is lowered in NIR.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17539>

radeonsi: Set vertex_id_zero_based

Let NIR lower load_vertex_id instead of LLVM.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17539>

v3dv: Use nir_gen_rect_vertices

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17535>

radv: Use nir_gen_rect_vertices

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17535>

nir: Add a common gen_rect_vertices implementation

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17535>

radv: disable viewport depth clamping only when necessary

When the application uses depth values outside of the [0.0,1.0] range
with VK_EXT_depth_range_unrestricted, or when explicitly disabled.
Otherwise, the driver can clamp to [0.0,1.0] internally for optimal
performance.

From the Vulkan spec "in 28.10.1. Depth Clamping and Range Adjustment":

    "If depth clamping is not enabled and zf is not in the range [0, 1]
    and either VK_EXT_depth_range_unrestricted is not enabled, or the
    depth attachment has a fixed-point format, then zf is undefined
    following this step."

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/16349>

v3dv: implement VK_KHR_buffer_device_address

This feature allows shaders to use pointers to buffers which may
not be bound via descriptor sets. Access to these buffers is done
via global intrinsics.

Because the buffers are not accessed through descriptor sets, any
live buffer flagged with VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_KHR
can be accessed by any shader using global intrinsics, so the driver
needs to make sure all these buffers are mapped by the kernel when
it submits the job for execution.

We handle this by tracking if any draw call or compute dispatch in
a job uses a pipeline that has any such shaders. If so, the job is
flagged as using buffer device address and the kernel submission
for that job will add all live BOs bound to buffers flagged with the
buffer device address usage flag.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17275>

broadcom/compiler: track if a shader uses global intrinsics

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17275>

broadcom/compiler: implement 2x32 global intrinsics

Notice we ignore the high 32-bit component of the address because
we know it must be 0.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17275>

nir: add nir_address_format_2x32bit_global

This adds support for global 64-bit GPU addresses as a pair of
32-bit values. This is useful for platforms with 32-bit GPUs
that want to support VK_KHR_buffer_device_address, which makes
GPU addresses explicitly 64-bit.

With the new format we also add new global intrinsics with 2x32
suffix that consume the new address format.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17275>