nir,spirv: Preserve inbounds access information

Preserving information about inbounds access and
the required bit size for the bounds will help
with avoiding 64-bit operations when lowering io.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16729>

dzn: Enable depth-bounds testing

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16948>

dzn: Use CreatePipelineState()

This way we can easily extend the logic to support new features, like
depth-bounds testing.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16948>

dzn: Drop an unused argument passed to dzn_graphics_pipeline_translate_vi()

The 'out' argument is not used, let's drop it.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16948>

dzn: Switch to ID3D12Device2

So we can use the CreatePipelineState method and unlock new features
like depth-bounds tests.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16948>

zink: always create descriptor templates

the 'notemplates' debug mode is somewhat misleading since there's no
uncached+notemplates mechanism, meaning that if the descriptor cache
explodes it'll still use templates for updating in the fallback path

Fixes: 4e3768914d4 ("zink: add ZINK_DESCRIPTORS env var to explicitly set a mode")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16927>

zink: fix buffer descriptor hashing

since all the buffers are now an array descriptor, hashing needs
to iterate over all the descriptors in order to be accurate

Fixes: a7327c7cac9 ("zink: implement indirect buffer indexing")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16927>

zink: simplify some compiler bo code

this is a bit easier to read

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16927>

Init macOS GitHub Action

Closes #6517

Signed-off-by: Yurii Kolesnykov <root@yurikoles.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16875>

Remove Travis CI config

Closes #4919

Signed-off-by: Yurii Kolesnykov <root@yurikoles.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16875>

v3d: add ARB_polygon_offset_clamp extension support

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16925>

panfrost: Constant stencil value tracking

If stencil is constant across the resource, then it can be treated as
if it was cleared.

Improves performance in applications which create a stencil buffer but
do not use it. Originally the same was done for depth to help some 2D
applications, but that gave mixed results so the patch was dropped.

v2: Don't do anything if a fragment job wouldn't be needed otherwise.
v3: Set stencil_value when a batch is cleared (Alyssa)
v4: Handle clears when the stencil is already known (Alyssa)
v5: Make sure shared resources are not used (Alyssa)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16646>

radv: advertise VK_EXT_primitives_generated_query

Pass all dEQP-VK.transform_feedback.primitives_generated_query.*.

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/15639>

radv: implement VK_EXT_primitives_generated_query

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/15639>

radv: add few helpers related to streamout

Streamout must be enabled for the PRIMITIVES_GENERATED query to work.

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/15639>

radv: flush the NGG query state when the argument is declared

When primitives generated query is used, the driver also needs to
emulate counting for NGG VS/TES.

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/15639>

ac/nir/ngg: count the number of generated primitives for VS and TES

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/15639>

radv: declare the NGG query argument for primitives generated query

To emulare counting for NGG VS/TES.

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/15639>

radv: add primitives_generated_query to the graphics pipeline key

Different shader variants will be used to not hurt when the features
aren't enabled.

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/15639>

radv: track if primitives generated query features are enabled

This will be used to conditionally declare the NGG query argument
for NGG VS/TES and also adjust the NGG shaders.

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/15639>

radv: rename ngg_gs_state to ngg_query_state

This state will be used by NGG VS and NGG TES too.

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/15639>

lima/ci: Don't skip piglit opengl 1 tests

Some extensions and features we support only get covered in this space.
We still get a 10min runtime, so it makes sense to include them from
now on.

Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16928>

svga: fix shared memory index

State tracker only declares one shared memory for GLSL and all
references to the shared memory are with index 0.
So fix the shared memory index in the declaration to use 0 index.

Fixes spec@arb_compute_shader@execution@shared*

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16930>

svga: fix reference to NULL offset for atomic counter

ntt does not supply an offset source for HWATOMIC.
So no need to reference the offset source for HWATOMIC.

Fixes spec@arb_compute_shader@execution@basic*

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16930>

glsl: inline do_common_optimization() call

The function was previously a helper for when some drivers still
called the GLSL IR optimisations in a loop. No drivers do that
anymore.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16924>

glsl: remove never true do_dead_code() parameter

Since we have now switched all drivers to using NIR and therefore
the NIR based uniform linker this param never needs to be set to
true so remove it.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16924>

docs: link releases notes for 21.3.9

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16931>

docs Add sha256 sums for 21.3.9

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16931>

docs: add release notes for 21.3.9

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16931>

nir/deref: Handle RESTRICT for SSBO deref bindings

Tested-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir/deref: Handle SSBO array bindings specially

Instead of just checking for the variables to match, check that the
entire deref up to the interface type matches.

Tested-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir/vars_tests: Use nir_var_mem_global instead of ssbo

We're about to add a bunch of SSBO special cases which will depend on
SSBOs always being either structs or arrays of structs.  All those
little vector SSBOs we're creating will no longer be valid.  Switch to
nir_var_mem_global to avoid this.

Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir: Increase nir_variable_data::mode to 16 bits

This is required if we want to have variables with nir_var_mem_global
which we will for CL eventually.  Also, they're useful in unit tests
because they're the most generic thing imaginable and can't get
eliminated by normal means.

Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir/deref: Re-arrange variable checks in compare_deref_paths

Instead of having a bunch of mode checks as special cases, assert that
the modes equal and then switch on the mode.  This should make the
special cases a bit easier to understand.  Handling of `a_var == b_var`
looks redundant now but it won't be in the next patch.

Tested-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir/deref: Make compare_deref_paths take a stop callback

This will let us use it to compare only the first part of a pair of
deref paths and continue the comparison later.

Tested-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir/deref: Use an index in compare_deref_paths

Instead of incrementing pointers, use an integer index.  This makes it
clear that we always increment them together.  It'll also make the next
change a bit easier.  We use a pointer to an integer because the next
patch is going to let us abort the walk and we want to be able to
continue where we left off.

Tested-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir/deref: Break out a helper for compare_deref_paths

Tested-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

nir/deref: stop assuming coherent accesses of different SSBOs may alias

Whether it's coherent should be irrelevant and the ACCESS_RESTRICT check
above should consider all cases aliasing unless NIR makes it clear they're
not.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Tested-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Cc: mesa-stable@lists.freedesktop.org
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16894>

radv, aco: Round texture array layer in NIR.

Foz-DB Navi21:
Totals from 9100 (6.75% of 134913) affected shaders:
VGPRs: 609912 -> 610104 (+0.03%); split: -0.01%, +0.05%
SpillSGPRs: 1459 -> 1489 (+2.06%)
CodeSize: 66705920 -> 66620288 (-0.13%); split: -0.13%, +0.00%
MaxWaves: 148546 -> 148518 (-0.02%); split: +0.00%, -0.02%
Instrs: 12278485 -> 12255821 (-0.18%); split: -0.19%, +0.00%
Latency: 277414916 -> 277261192 (-0.06%); split: -0.08%, +0.02%
InvThroughput: 48431180 -> 48394637 (-0.08%); split: -0.11%, +0.03%
VClause: 250866 -> 251062 (+0.08%); split: -0.04%, +0.11%
SClause: 498377 -> 498173 (-0.04%); split: -0.08%, +0.04%
Copies: 652835 -> 655371 (+0.39%); split: -0.09%, +0.48%
Branches: 284367 -> 284371 (+0.00%); split: -0.00%, +0.00%
PreSGPRs: 498580 -> 498477 (-0.02%)
PreVGPRs: 558436 -> 558709 (+0.05%); split: -0.01%, +0.06%

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16871>

nir/lower_tex: Add lower_array_layer_round_even option.

Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16871>

radv: Advertise VK_KHR_ray_tracing_maintenance1

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

radv: vkCmdTraceRaysIndirect2KHR

This changes the trace rays logic to always use
VkTraceRaysIndirectCommand2KHR and implements
vkCmdTraceRaysIndirect2KHR. I renamed the
load_sbt_amd to sbt_base_amd and moved the SBT
load lowering from ACO to NIR.

Note that we can not just upload one pointer to
all the trace parameters because that would
be incompatible with traceRaysIndirect.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

nir: Handle ray_launch_size_addr in opt_preamble

Found this while working on traceRaysIndirect2.
I don't think this is relevant for now at least
since we don't use the pass in RADV.

Fixes: 938c9d9 ("nir: Add a ray launch size addr intrinsic")
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

radv: PIPELINE_STAGE_2_ACCELERATION_STRUCTURE_COPY

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

radv: ACCESS_2_SHADER_BINDING_TABLE_READ_BIT_KHR

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

radv: Add accel struct queries for maintenance1

Implements the new
VK_QUERY_TYPE_ACCELERATION_STRUCTURE_SERIALIZATION_BOTTOM_LEVEL_POINTERS_KHR
and
VK_QUERY_TYPE_ACCELERATION_STRUCTURE_SIZE_KHR
query types.

The documentation is a bit lacking for now but
the fist type probably refers to the instance
count and the second type refers to the
acceleration structure size which we already
store in radv_acceleration_structure. To support
size queries, this commit adds a size member
to the acceleration structure header.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

radv: ray cull mask

Lower the cull_mask system value to rt variable
load and advertise the corresponding capability.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

spirv: Add plumbing for ray_cull_mask

Add a new cull_mask system value that is exposed
by the ray_cull_mask capability of
SPV_KHR_ray_cull_mask.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

vulkan: Spec update to 1.3.216

Update the vulkan headers and xml to 1.3.216
including the spirv headers. The new spirv spec
added new OpAlias*INTEL ops, which we ignore in
vtn_gather_types_c.py.

Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16430>

u_transfer_helper: flush temporary transfer first for MSAA

vtbl->transfer_map may return staging buffer and not real one and it
exposes a problem in MSAA resolve path, since u_transfer_helper does
blit from a resource that is still mapped and it's not flushed yet.

Add explicit flush_region() for a temporary transfer before doing flush
for MSAA resolve.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16923>

aux/tc: handle link_shader

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16906>

aux/trace: add pipe_context::link_shader

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16906>

panvk: Drop support for Midgard

We've discussed this at length and have agreed that Midgard + Vulkan is DOA, but
have let the code linger. Now it's getting in the way of forward progress for
PanVK... That means it's time to drop the code paths and commit t to not
supporting it.

Midgard is only *barely* Vulkan 1.0 capable, Arm's driver was mainly
experimental. Today, there are no known workloads today for hardware of that
class, given the relatively weak CPU and GPU, Linux, and arm64. Even with a
perfect Vulkan driver, FEX + DXVK on RK3399 won't be performant.

There is a risk here: in the future, 2D workloads (like desktop compositors)
might hard depend on Vulkan. It seems this is bound to happen but about a decade
out. I worry about contributing to hardware obsolescence due to missing Vulkan
drivers, however such a change would obsolete far more than Midgard v5...
There's plenty of GL2 hardware that's still alive and well, for one. It doesn't
look like Utgard will be going anywhere, even then.

For the record: I think depending on Vulkan for 2D workloads is a bad idea. It's
unfortunately on brand for some compositors.

Getting conformant Vulkan 1.0 on Midgard would be a massive amount of work on
top of conformant Bifrost/Valhall PanVK, and the performance would make it
useless for interesting 3D workloads -- especially by 2025 standards.

If there's a retrocomputing urge in the future to build a Midgard + Vulkan
driver, that could happen later. But it would be a lot more work than reverting
this commit. The compiler would need significant work to be appropriate for
anything newer than OpenGL ES 3.0, even dEQP-GLES31 tortures it pretty bad.
Support for non-32bit types is lacklustre. Piles of basic shader features in
Vulkan 1.0 are missing or broken in the Midgard compiler. Even if you got
everything working, basic extensions like subgroup ops are architecturally
impossible to implement.

On the core driver side, we would need support for indirect draws -- on Vulkan,
stalling and doing it on the CPU is a nonoption. In fact, the indirect draw code
is needed for plain indexed draws in Vulkan, meaning Zink + PanVK can be
expected to have terrible performance on anything older than Valhall. (As far as
workloads to justify building a Vulkan driver, Zink/ANGLE are the worst
examples. The existing GL driver works well and is not much work to maintain. If
it were, sticking it in Amber branch would still be less work than trying to
build a competent Vulkan driver for that hardware.)

Where does PanVK fit in? Android, for one. High end Valhall devices might run
FEX + DXVK acceptably. For whatever it's worth, Valhall is the first Mali
hardware that can support Vulkan properly, even Bifrost Vulkan is a slow mess
that you wouldn't want to use for anything if you had another option.

In theory Arm ships Vulkan drivers for this class of hardware. In practice,
Arm's drivers have long sucked on Linux, assuming you could get your hands on a
build.  It didn't take much for Panfrost to win the Linux/Mali market.

The highest end Midgard getting wide use with Panfrost is the RK3399 with the
Mali-T860, as in the Pinebook Pro. Even by today's standards, RK3399 is showing
its limits. It seems unlikely that its users in 10 years from now will also be
using Vulkan-required 2030 desktop environment eye candy. Graphically, the
nicest experience on RK3399 is sway or weston, with GLES2 renderers.
Realistically, sway won't go Vulkan-only for a long-time.

Making ourselves crazy trying to support Midgard poorly in PanVK seems like
letting perfect (Vulkan support) be the enemy of good (Vulkan support). In that
light, future developers making core 2D software Vulkan-only (forcing software
rasterization instead of using the hardware OpenGL) are doing a lot more
e-wasting than us simply not providing Midgard Vulkan drivers because we don't
have the resources to do so, and keeping the broken code in-tree will just get
in the way of forward progress for shipping PanVK at all.

There are good reasons, after all, that turnip starts with a6xx.

(If proper Vulkan support only began with Valhall, will we support Bifrost
long term? Unclear. There are some good arguments on both sides here.)

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Acked-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16915>

lavapipe: allow null handleTypes

handleTypes is allowed to be null with VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO

Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Closes #6501

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16590>

microsoft/compiler: Drop the auto_link parameter passed to nir_to_dxil()

Now that all nir_to_dxil() have transitioned to explicit linking, we
can drop the auto_link parameter and simplify the signature logic along
the way.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/spirv_to_dxil: Let spirv_to_dxil() handle the linking

That's still a dummy linking we do here, and it's likely to be
incorrect for complex scenarios. Not sure the previous situation
was any better though, and at least, doing that allows us to
get rid of the special cases we have in dxil_signature.c.

Reviewed-by: Enrico Galli <enrico.galli@intel.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

dzn: Fix vertex input handling

Vertex shaders are allowed to define input variables pointing to the
same location but a different, or even variables that overlap other
variables, as long as only one of them is used in a shader invocation.

One way to support that case would be to merge overlapping variables,
but we can also declare one input element per variable, and make those
point to the same input slot/offset. The only limitation with the
second approach is the maximum number of VS input registers, meaning
that only (32 - num_sysvals) input variables can be defined.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

dzn: Fix unitialized var access in dzn_meta_compile_shader()

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

dzn: Handle varying linking explicitly

Now that dxil_spirv_nir.h exposes an helper to run the
DXIL-SPIRV specific passes, we can handle the varying linking
on our side and tell nir_to_dxil() we don't want automatic
varying index/register assignment, which should fix a bunch
of compiler errors.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

dzn: Fix maxVertexInputAttributes limit

The current implementation doesn't allow more than
MAX_VERTEX_GENERIC_ATTRIBS vertex attributes.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/spirv_to_dxil: vectorize IOs

DXIL doesn't like when 2 different variables occupy the same slot, so
let's vectorize IOs to avoid that.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/compiler: Allow vulkan users to handle varyings linking

Letting the compiler decide which slot should be used for varyings when
it doesn't know about the varyings written/read by the previous/next
stage doesn't work well. So let's the caller decide when it wants
automatic index/register assignment through a dedicated parameter,
instead of assuming Vulkan users always want that.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/spirv_to_dxil: Add a linking helper

Linking should be done in reverse order, starting from the last
pipeline stage and going backward, so we can eliminate outputs from the
previous stage that are never used by the next stage, and possibly
kill some instructions and input variables too.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/spirv_to_dxil: Remove dead IO vars at the end of dxil_spirv_nir_passes()

No need to keep them around if they're unused. Moreover, this should
allow the linking step to get rid of outputs when the next stage
doesn't use them.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/compiler: Make input_mappings immune to driver_location holes

Right now, nir_to_dxil() assumes driver_location on inputs will be
contiguous, which is true for GL, and also true for Vulkan shaders
with the current implementation. But we are trying to delegate
the varying linking step to Dozen, and that means the driver will
assign the driver_location field.

For everything except vertex shaders this works fine, because we
are in control of the ID we assign to each variable, and can make
sure no holes exists in this assignment, but vertex inputs expect
the index value (which is directly extracted from the
driver_location field) to match the input index defined at pipeline
creation time. The compiler has a hack to treat Vulkan differently
and extract the index from the var->data.location field instead,
but that's a bit confusing.

Moreover, the input_mappings[] array is already indexed with
the var->data.driver_location field in the input load emission
path, so it makes sense to index it with the same field when
emitting signatures.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/spirv_to_dxil: Constify the descriptor binding mapping fields

spirv_to_dxil() doesn't modify those, so let's constify them.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/spirv_to_dxil: Extract NIR passes out of spirv_to_dxil()

Extract NIR passes out of spirv_to_dxil() so we can re-use them
without separately and do the varying linking in Dozen. This way
we will also be able to use vk_shader_module_to_nir() which
takes care of the SPIRV -> NIR translation, plus a bunch of
common lowering passes.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/compiler: Make sure all SV_Position components are written

Not entirely sure if writing to only some components of
gl_Position/SPIRV-PositionBuiltin from a vertex-processing stage is
allowed, but
https://github.com/dneto0/spirv-samples/blob/main/spvasm/SpvModuleScopeVarParserTest_BuiltinPosition_StorePositionMember_OneAccessChain.spvasm
does exactly that.

Let's make sure all SV_Position components are written, otherwise the
DXIL validator complains.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

microsoft/spirv2dxil: Add a debug option

So we can dump NIR shaders.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

ci/windows: Bump spirv-samples commit-ID

We need the fix adding a Block decoration to the BuiltIn struct in
SpvModuleScopeVarParserTest_BuiltinPosition_BuiltIn_Position_Initializer.spvasm.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16221>

gitlab: ask that reporters don't include long logs in descriptions

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14733>

gitlab: ask for useful titles

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14733>

asahi,agx: Don't set PIPE_CAP_POINT_COORD_ORIGIN_UPPER_LEFT

Tell the state tracker our point coordinates have a lower left origin
instead of an upper left origin, and remove our point coordinate
flipping code. Saves an instruction in any shader that reads
gl_PointCoord.y

Note: the OpenGL blob also emits an "fadd $y', ^y.neg, 1.0" to flip
point coordinates, so this isn't just a Metal weirdness.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16829>

mesa,gallium: Make point coord origin a CAP

When lower_wpos_pntc is used, the state tracker inserts code to
transform gl_PointCoord.y according to a uniform, to account for
API-requested point coordinate origin and framebuffer orientation. With
the transformation, driver-supplied point coordinates are expected to
have an upper left origin.

If the hardware point coordinate supports (only) a lower left origin,
the backend has to use lower_wpos_pntc and then lower *again* to flip
back. This ends up transforming twice, which is wasteful:

   a = load point coord Y with lower left origin
   a' = 1.0 - a
   a'' = uniform_transform(a')

However, lower_wpos_pntc is quite capable of transforming for a lower
left origin too, it just needs to flip the transformation. Add a CAP
specifying the point coordinate origin convention, rather than assuming
upper-left. This simplifies the Asahi code greatly.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16829>

pan/perf: Fix performance counters on G57

The performance counter layout depends on the number of L2 blocks and the number
of shader cores. It doesn't make a ton of sense to hardcode these into the XML
files. Instead, let's make the coutner offsets in the XML files relative to the
categories (blocks), so we can calculate the offsets of the categories
themselves at runtime based on the computed layout. This fixes performance
counters on Mali-G57 as implemented on MT8192.

There is little code change here, mainly churn from changing the XML definition.
Postprocessing for the XML to make it suitable for Mesa uses Antonio Caggiano's
https://gitlab.freedesktop.org/panfrost/hwc-helper tool.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Antonio Caggiano <antonio.caggiano@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16803>

panfrost: Add panfrost_query_l2_slices helper

The number of L2 performance counter blocks equals the number of L2 slices, so
add a query to get this. This information isn't needed by the Mesa driver, so
don't get it in the default device initialization path.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Antonio Caggiano <antonio.caggiano@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16803>

panfrost: Remove is_64b assignments

These are redundant with GenXML defaults, they're just noise.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16916>

panfrost: Remove redundant first_tag access

This already happens in the common prepare_rsd call.

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/16916>

panfrost: Deduplicate indirect dispatch structs

The input is specified in two identical structs, tear that apart.

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/16916>

panvk: Use push constants for copy shaders

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/16916>

panvk: Use push constants for clear

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/16916>

panfrost: Use push constants for indirect draws

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/16916>

panfrost: Use push constants for indirect dispatch

Much simpler than creating a UBO and relying on it getting optimized to a push
constant, with possible reordering.

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/16916>

pan/bi: Don't reorder push with no_ubo_to_push

Otherwise, load_push_constant won't work properly. This could probably be made
to work if we tried hard enough, but we still don't want reordering for internal
(meta) shaders which are layed out deliberately.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16916>

pan/bi: Implement load_push_constant

Bifrost supports "fast access uniforms" loaded from a single contiguous buffer.
This maps directly to Vulkan push constants, with some caveats:

* No indirect access. Indirects need to be lowered to a UBO pull.
* Strict alignment requirements. These will be met in practice.

Implement the NIR intrinsic and map it to the native hardware construct.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16916>

panfrost: Fix inverted indirect dispatch condition

Inverted condition -- indirect dispatch gets disabled when WLS is in use, not
the other way around. Not sure how this worked before...

Fixes: fd7b44882ce ("panfrost: Use direct dispatch with shared memory")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16916>

turnip: consider render pass costs in autotune

To be able to sum drawcall cost and render pass cost, the units of costs
are changed to bytes.  With that, tu_autotune_use_bypass can make
decisions by comparing the costs of sysmem rendering and gmem rendering.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16733>

turnip: estimate render pass costs

They will be used by autotuner.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16733>

turnip: include stencil test for drawcall costs

Stencil test reads from and writes to the stencil buffer.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16733>

turnip: base pipeline draw costs on colorWriteMask

drawcall_base_cost estimates the memory bandwidth per sample.
colorWriteMask should be more accurate than fs outputs.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16733>

turnip: if-checks autotune debug macros

This avoids bitrot while the compiler can easily optimize away those
checks.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16733>

radeonsi: set the max UBO size same as the max SSBO size

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16901>

radeonsi: compute PIPE_CAP_MAX_TEXEL_BUFFER_ELEMENTS_UINT correctly

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16901>

radeonsi: change max TBO/SSBO sizes again and rework max alloc size

Allow 1/4 of the max heap size, but maximum of 512 MB on 32-bit
architectures.

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16901>

radeonsi: clamp against MAX_TEXEL_BUFFER_ELEMENTS correctly

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16901>

radeonsi: report correct maximum compute grid sizes

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16901>

radeonsi: increase the max compute LDS size to 64KB for gfx7+

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16901>

pvr: debug: Implement PVR_DEBUG variable

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16907>

pvr: csbgen: Add *_unpack() functions for all generated struct types

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16884>

pvr: csbgen: Fix "local variable may be undefined" lints

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16884>