pvr: debug: Enhancements to hex dumps

Contains the following enhancements & fixes:
 - Increase (decrease?) the granularity to single bytes rather than
   using an arbitrary word size,
 - Remove some spurious semicolons at the end of macros, and
 - Do not collapse sections of zero bytes that consist of only a single
   line.

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20040>

pvr: Remove a todo in vkCmdEndRenderPass2().

The first end_sub_cmd() is to make sure that we end the last sub_cmd.
The end_sub_cmd() in pvr_resolve_unemitted_resolve_attachments() makes
sure that we end any transfer sub_cmds created in there.

Suggested-by: Frank Binns <frank.binns@imgtec.com>
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19957>

pvr: Replace sub_cmd flags with bools within each sub_cmd type.

This commit remove:
 - PVR_SUB_COMMAND_FLAG_TRANSFER_SERIALIZE_WITH_FRAG.
 - PVR_SUB_COMMAND_FLAG_OCCLUSION_QUERY.

The first flag was specific to transfer sub commands and the last
one, for graphics ones. Now we just have a bool in the transfer
sub_cmd, and one in the graphics sub_cmd.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19957>

pvr: Handle PVR_SUB_COMMAND_FLAG_OCCLUSION_QUERY.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19957>

pvr: Handle PVR_SUB_COMMAND_FLAG_TRANSFER_SERIALIZE_WITH_FRAG.

The flag was previously named PVR_SUB_COMMAND_FLAG_WAIT_ON_PREVIOUS_FRAG.
Since the next fragment job is also made to wait for the transfer
job to complete, the previous name might have been a bit misleading.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19957>

v3dv: fix job serialization for single sync path

The idea in the single sync path is that we serialize any job that
needs to wait, however, our ANY queue syncobj only tracks the last job
submitted to any hardware queue, so in practice when we wait on this
we are only serializing against the queue to which we have submitted
the last job, which is not correct.

Fix that by accumulating the last job sync into the ANY queue synbcobj
to ensure that waiting on this syncobj effectively waits on all
hardware queues.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20078>

v3dv: make single-sync paths more explicit

Instead of having functions that return early in multi-sync mode
let's only call them when we are in single-sync mode. I think this
makes the code more explicit.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
cc: mesa-stable

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20078>

panfrost: Add NATIVE_FENCE_FD cap

Add support for NATIVE_FENCE_FD so panfrost can advertise support for
EGL_ANDROID_native_fence_sync.

Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19774>

panfrost: Move fence code to pan_fence.{c,h}

Before adding support for NATIVE_FENCE_FD, let's move the fencing logic
to a dedicated file to avoid spreading the code in different places.

Suggested-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19774>

panfrost: Destroy panfrost_context::syncobj in the ctx desctruction path

Destroy panfrost_context::syncobj in the ctx desctruction path so we
don't leak a sync object.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19774>

ac/nir/ngg: rename nogs 16bit output mask and var

To represent 16bit outputs more clearly.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19697>

ac/nir/ngg: gs support 16bit outputs

radeonsi uses 16bit varying slots.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19697>

ac/nir/ngg: gs skip check bit size before nir_u2u

nir_u2u do for us.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19697>

ac/nir/ngg: gs store output use src_type index for type info

More precise type info, can be used for 16bit output streamout
to convert 16bit int/uint/float to 32bit one later.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19697>

ac/nir/ngg: gs use u_foreach_bit64 to loop all output slots

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19697>

ac/nir/ngg: reduce nogs 16bit output gather space

Max slot number for 16bit output is 16, so no need to use
64 array size for them.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19697>

util/dynarray: Add an append_array helper

Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19685>

intel/compiler: user payload starts after TUE header & its padding

All data written by the user are offset by TUE header size.
Without this patch we copy the correct amount of user data, but both
"from" and "to" offsets are wrong.

Fixes: 37e78803d7b ("intel/compiler: use nir_lower_task_shader pass")

Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19409>

nir/lower_task_shader: allow offsetting of the start of payload

We need this, because on Intel task payload starts with private header,
followed by user-accessible data.

Fixes: 37e78803d7b ("intel/compiler: use nir_lower_task_shader pass")

Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19409>

intel/compiler: adjust [store|load]_task_payload.base too

Base also needs to be converted from bytes to words.

Fixes: c36ae42e4cc ("intel/compiler: Use nir_var_mem_task_payload")
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19409>

ci/zink: add lavapipe flakes

Listed from: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7613
Bug: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7781

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20093>

zink: do not complain about missing line-stipple support

We can lower this now, so let's not complain about it...

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

zink: lower line stipple

This lowers line-stippling to a combination of geometry and fragment
shaders:

- The geometry shader computes the length of each line-segment, and
  outputs a varying that produces the stipple position.
- The fragment shader looks up the stipple position in the
  stipple-pattern once per sample, and updates the sample mask
  accordingly.

In case there's no geometry shader in place, we create a new
pass-through shader.

We should probably not declare the the push-constants in the pipeline
layout unless they're actually needed. But we already do this
unconditionally for the vertex shader and tesselation push-constants, so
let's do it unconditionally for these as well for now.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

zink: allow to generate any vertex shader stage

There's times when it's going to be useful to generate geometry shaders
as well, so let's generalize the infrastructure for generated shader
stages a bit.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

zink: process non-optimal-key passes first

Right now, it's only the vertex-shader that needs special handling for
non-optimal keys. That makes it possible to use fallthrough to always
end up in the last-vertex-stage conditional.

But we're about to add special handling for the geometry stage as well,
so let's prepare by splitting the switch-statement in two; one that only
happens for non-optimal keys, and does all the needed processing there,
and one that deals with the rest.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

zink: give gs its own shader-key

Line-stipple lowering is going to need some geometry-shader specific
lowering, so lets give the GS its own shader-key struct.

The GS variant only needs a non-optimal variant, so let's assert that to
be sure.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

zink: emit vars with nir_var_shader_temp mode

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

zink: add line-stippling lowering passes

There's two notable limitations here:
- This will viewport-map to viewport #0 only. This is because we need
  the viewport-scale factors, which we'll be uploading using
  push-constants. And we don't want to waste too many of those...
- It's missing a "global" stipple-counter. It doesn't seem like there's
  a portable way of implementing this, so this is going to require a VK
  extension that can be implemented in a hardware-specific way in the
  long run. For now, let's just ignore the global stipple counter.

These two limitations don't seem viable to overcome for now, so but this
is better than nothing.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

zink: setup driver-workaround for missing linestipple

This is not ideal, but at least it should work. In the long run, we
might want to store a bit per mode we're missing, so we can do this
conditionally. But that's quite a bit more complicated, so let's go with
this for now.

The line-stippling logic needs non-optimal shader-keys. So let's drop
some perf on the floor here.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>

ci/alpine: disable the job, still occasionally flakes

See: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/32689466

Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20076>

glx: Require __DRI_SWRAST >= 4 for doing swrast.

The only implementer in tree is v4.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

egl: Refactor common error handling for context creation.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

egl: Bump minimum version of __DRI_SWRAST to 4.

The only implementer in tree is v4.  This simplifies some bits now that we
always have CreateContextAttribs.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

glx: Require __DRI_DRI2 v2 for doing X11 DRI2.

The only implementer in tree is v4.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

egl: Require __DRI_DRI2 v4 if we're to do DRI2.

The only in-tree implementers are v4.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

egl: Bump __DRI_CONFIG_OPTIONS min version to 2.

The only implementer in tree is v2.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

egl: Bump minimum version of __DRI_IMAGE to 6 and drop version checks.

All __DRI_IMAGEs in tree are v6+ (lowest being drisw) and implement
createImageFromTexture.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

egl: Bump required version of optional __DRI2_FENCE to 2.

The only implementer of it in tree is v2, so no need for checks.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

gbm: Bump required __DRI_IMAGE version to 6.

That's the minimum implemented in tree (gallium swrast).  Drops a few more
version checks.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

gbm: Drop support for __DRI_DRI2 < 4 and __DRI_SWRAST < 4.

We're always loading a gallium driver built from this tree, so it's always
v4.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

dri: Add notes on what part of the loader interface are used by Xorg.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20044>

freedreno/fdl: Set sRGB bit for storage images

This probably wasn't noticed earlier because tests using sRGB storage
images didn't exist, and we didn't know whether this works, but this
fixes dEQP-VK.image.store.without_format.2d.*_srgb which also proves
that the bit works.

Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20060>

agx: Clamp point sizes

Fixes vs-point_size-zero.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

agx: Handle 32-bit gl_FragCoord.zw

The coefficient register is 16-bit so our builder will make the iter 16-bit too
(maybe not the best design...), force fp32 to match the NIR intrinsic.

Fixes glsl-fs-fragcoord-zw-ortho

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

agx: Handle large varying indices

Fixes glsl-max-varyings.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

asahi: Support one-sided polygon modes

We can implement glPolygonMode(GL_FRONT_AND_BACK, ...) natively. What we can't
implement natively are two-sided polygon modes. For that Apple has a nontrivial
lowering which I don't feel the need to implement unless someone actually hits a
workload other than Piglit that uses it.

Vulkan requires only one-sided polygon modes (so this is sufficient there), and
GLES doesn't have polygon modes at all. If an app hits the unimplemented case,
throw a warning like Zink does.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

asahi: Handle NULL sampler views

Fixes fp-fragment-position (crash->pass).

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

asahi: Unset GL_CLAMP

Use the Zink lowering for the legacy mode, it's not too many instructions on AGX
anyway. Fixes texwrap tests.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

asahi: Set frag coord caps correctly

Fixes ./glsl-arb-fragment-coord-conventions, c.f. 12facf23b1f ("panfrost: Don't set CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER").

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

asahi: Set PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION

Fixes arb-provoking-vertex-render.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

asahi: Set PIPE_CAP_SUPPORTED_PRIM_MODES

To lower GL_POLYGONS which we don't do natively. Fixes a pile of crashes in
Piglit.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>

hasvk: Drop anv_nir_add_base_work_group_id()

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

anv: Drop anv_nir_add_base_work_group_id()

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

intel/nir: Set has_base_workgroup_id for lower_compute_system_values

This option didn't exist half a decade ago when I first implemented base
workgroup support in ANV.  It's cleaner to just have split system values
like all the other zero_base+base things do.

We currently only do this for COMPUTE and not KERNEL because it lets us
avoid changing intel_clc for now.  We can add KERNEL later if needed.
We also don't do this lowering for task/mesh.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

hasvk: Implement lower_base_workgroup_id

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

anv: Implement lower_base_workgroup_id

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

crocus: Lower load_base_workgroup_id to zero

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

iris: Lower load_base_workgroup_id to zero

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

intel/fs: Support load_workgroup_id_zero_base

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

nir/divergence: Handle base_workgroup_id and workgrpu_id_zero_base

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>

radv: Remove the old LBVH shader.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19891>

radv: Switch to new LBVH implementation.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19891>

radv: Add new LBVH shaders.

Contrary to the previous implementation, this actually implements an LBVH builder.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19891>

radv: Handle nodes with 2 invalid children in internal node converter.

Fixes: 682dc5c28e4 ("radv: Add conversion shader for internal nodes")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19891>

panfrost: Enable AFBC of more formats

Enable AFBC for all RGBA UNORM formats possible in v5. This does not
cover the AFBC rules for newer gens, nor for YUV.

Noticed with an uncompressed R8 UNORM texture in SuperTuxKart.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19758>

panfrost: Enable AFBC of sRGB formats

AFBC of sRGB formats should just work. We just need to flip it on and enjoy
the improved performance.

In particular, this means that RGBA8 UNORM and RGBA8 sRGB UNORM are now
considered compatible formats for AFBC. That's a bug fix, because
GALLIUM_HUD use will act like a texture view between sRGB and linear
views. For FBOs, that will "just" result in a decompression, hurting
performance. For window system rendering with AFBC, that will cause an
assertion failure, as we cannot decompress SHARED resources.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19758>

panfrost: Enable AFBC of cube maps

Missed by mistake. This is not the same as 3D AFBC, it's just like a 2D
array. Noted in a supertuxkart pandecode.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19758>

panfrost: Handle all RGB AFBC modes on v9

We're about to enable AFBC on more formats in the core AFBC code. The plane
descriptor packing needs to be aware of these new formats.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19758>

aco: Use wave size specific opcode for s_or in cube map coord code.

Cc: mesa-stable
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20041>

nir/builder: Also short-circuit for auto-generated nir_t2t<N>()

This makes nir_i2i32(b, x) behave exactly like nir_i2iN(b, x, 32) etc.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7787
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>

nir/builder: Rework the boolean conversion helpers

Move them up to where the other conversion helpers.  For nir_b2<T>(),
suffix them with N like all the others and make them use
nir_type_convert() as well.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>

nir/builder: Drop nir_i2i and nir_u2u in favor of nir_x2xN

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>

nir/builder: Move conversions higher in nir_builder.h

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>

nir/builder: Short-circuit in nir_type_convert if no conversion happens

If both types are the same or both are integer types with the same bit
size, no actual conversion happens and nir_type_conversion_op() will
return nir_op_mov.  In this case, there's no point in emitting the move
and we can just return src instead.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>

nir/builder: Fix indentation of nir_type_convert

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>

nir: Fix builder usage in lower_mediump_vars()

In our handling of load_deref, we were calling builder helpers to create
conversions and then adjusting the destination bit size of the load.  We
should adjust the bit size first because the builder sometimes looks at
the bit sizes of SSA values passed in as arguments.

Even though it's not strictly necessary, adjust the store_deref case as
well to make it fully symmetric with the load_deref case.

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>

pan/mdg: Emulate 8-bit with the 16-bit pipe

We don't care to support i8vec16, we just need a bit of 8-bit support to
implement format packing/unpacking in blend shaders. We're already doing
this by using the 16-bit pipe, we just need to commit to it all the way
-- reporting the correct sizes in max_bitsize_for_alu so the mask
packing logic works as intended -- and dropping the imov-specific hack
that was introduced to workaround a similar class of bugs.

With the previous patch, fixes:

   dEQP-GLES31.functional.draw_buffers_indexed.random.max_required_draw_buffers.1

Fixes: 39e4b7279dc ("pan/midg: Fix swizzling on 8-bit sources")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19763>

pan/mdg: Refuse to schedule CSEL.vector to SMUL

Even if we only mask a single component from the result of CSEL.vector,
in our IR we treat its semantics as vector which causes trouble with
when scheduled to a scalar unit.

The problematic bundle looks like this:

   vmul.MOV.i32 R31, TMP0.xxxx, R0.yzww
   sadd.MAX.i32 TMP0.y, R0.y, #65408
   smul.CSEL.vector.i32 R0.y, TMP0.y, #127

As the comment in midgard.h illuminates, these CSEL instructions are
actually operating per-bit, lining up with the all-1's booleans in
Midgard. The Bifrost analogue is MUX.i32.bit, not CSEL.i32. We should
probably rename the Midgard instruction to make that clear.

Anyhoo, on the scalar unit, CSEL/MUX operates on the bottom 32-bits of
its source. That's ok for the usual r31.w case, because that's secretly
replicating to its nonexistent register, I think? But that doesn't work
with the CSEL.vector (MUX.vector) form, because the condition it's
actually muxing on is r31.x, which here is R0.y, not the intended R0.x.

Rather than adding more special cases to the already overcomplicated
scheduler (for the dubious benefit of avoiding a small shaderdb
regression), just avoid scheduling CSEL.vector to smul.

With the next patch, fixes:

dEQP-GLES31.functional.draw_buffers_indexed.random.max_required_draw_buffers.1

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19763>

ci: Rebalance radv/grunt testing

We've recently rebalanced our lab devices to get a fewer number of
grunts. Switch to scheduling only on the newer shinier ones, running
fewer tests. We'll evaluate the runtime, and if they're quick enough
then we can increase the amount of testing we do.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20081>

pvr: Add support to submit occlusion query sub cmds.

Co-authored-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Create a separate compute context for queries.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Set isp userpass to dirty for secondary command buffers.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Process queries and emit write program when ending sub_cmd.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Kick job and end sub_cmd if barrier_store is set in vkCmdExecuteCommands.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Process secondary buffer queries in vkCmdExecuteCommands.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Save error result in cmd_buffer state.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Implement vkCmdCopyQueryPoolResults API.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Implement vkCmdResetQueryPool API.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Add support to generate query programs.

This commit adds support to generate three types of query related
programs. PVR_QUERY_TYPE_AVAILABILITY_WRITE allows to submit the
queries, PVR_QUERY_TYPE_RESET_QUERY_POOL allows to reset the pool
and PVR_QUERY_TYPE_COPY_QUERY_RESULTS is to copy the results.

Co-authored-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Add PVR_SUB_CMD_TYPE_OCCLUSION_QUERY type sub cmd.

Co-authored-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Create device specific compute query programs.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Implement vkCmdEndQuery API.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Implement vkCmdBeginQuery API.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Sequential dependency should be NONE for 0 constant shared regs.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Implement vkGetQueryPoolResults API.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Add support to generate update compute kernel.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Add support to generate compute kernel to update shared regs.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

pvr: Change CDM to compute, TA to geometry and 3D to fragment.

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19371>

r600: Store nir shaders serialized to save memory

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7247

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20061>

aco: Don't prematurely emit s_andn2.

Split s_not + s_and allows more inverse comparision and s_cbranch_vccz
optimizations.

Foz-DB Navi21:
Totals from 516 (0.38% of 134913) affected shaders:
CodeSize: 7273724 -> 7273720 (-0.00%)
Instrs: 1364408 -> 1364407 (-0.00%)
Latency: 14604862 -> 14604858 (-0.00%)

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19143>