zink: only update gfx pipeline cache after creating a real pipeline

async pipelines may not require updates here

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12842>

zink: remove some ctx references from shader/pipeline compile

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12842>

zink: remove ctx references from shader compile path

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12842>

zink: make tcs shader generation take screen param

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12842>

zink: move pending prim type to gfx pipeline struct

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12842>

llvmpipe/cs: rework coroutine context handling (v2)

Get comfy.

llvmpipe coroutines have a stack frame. This is created by hooking
in malloc and coro.alloc and coro.size intrinsics.

LLVM has an CoroElide pass that is meant to allow that stack frame
to be done as an alloca in the caller instead of using the malloc path.

The CoroElide pass relies on the coroutine being inlined (fixed that).

The CoroElide pass relies on there being a direct connect between
coro.destroy(i8 *arg) and arg = coro.begin(id). However due to the
way the compute shaders are launched, there is no way to ensure that
link. Fixing the CoroElide pass seems quite difficult, I considered
having a force CoroElide always flag to make it dtrt, however I'm not
sure how ugly that would end up.

My first attempt tried to preallocate the stacks at a fixed size,
this turned out to be naive as the stack frame size was not sized
like I expected. Instead the first coro to run allocs enough for
everyone, so avoid the massive amounts of small allocations.

This remove coro malloc from a lot of profiles and shaves another 30s
or so from OpenCL ./conversions/test_conversions uchar_uin
(from 4.40m to just under 4m on my ryzen 7 1800x)

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12432>

llvmpipe: shorten hold time on the screen mutex

There is no requirement to hold this mutex over the wait. I doubt
it matters much in practice.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12432>

gallivm/coro: use a phi instead of alloca

this just matches what the docs recommend

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12432>

llvmpipe/cs: rework thread pool for avoid mtx locking

This helps reduced the mtx lock/unlock overheads for the threadpool
if the work evenly distributes across the number of threads.

The CL CTS conversions tests really hit this, and this takes maybe 10-20s
off a 5min test run.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12432>

zink: fix enabled vertex buffer mask calculation

the mask can't entirely be calculated based on the integer parameters,
as it's possible for some of the "bind" slots to actually be unbinds,
so remove bits as necessary to fix this

also add some debug asserts to ensure I don't break this again for the
tenth time

Fixes: 6dd02a5139a ("zink: stop using util_set_vertex_buffers_mask()")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12871>

pan/bi: Don't set dependencies for +BLEND in blend shaders

The dependency wait should already have been done in the fragment
shader.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12461>

lavapipe: enable dynamic index ubo/ssbo

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12689>

gallivm/ssbo: cast ssbo index to int type.

Since these can be loaded from ubos or other places now.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12689>

gallivm/ssbo: fix up dynamic indexed ssbo load/stores/atomics

Although the index has to be dynamically uniform, if we don't ever
execute a few lanes then we'll have 0, so it important to read the
ssbo index from the first active lane.

Just loop over them all.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12689>

docs: shorten "last release" note to fit on the website without horizontal scrolling

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12798>

docs: add plan for 21.3.x release cycle

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12397>

egl: Add a basic Windows driver

This is enough to run wglgears ported to EGL (desktop GL)

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

egl: Detect Windows platform using GDI

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

egl: Don't try to dereference native displays unless there's a detectable platform

Not all native displays are pointers to dereferenceable memory, e.g.
DCs on Windows. Don't bother dereferencing if no platforms are available
that can be detected that way.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

egl: Use the .def file for Windows

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

egl: Add wgl/gallium dependencies for Windows platform

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

meson, egl: Support building for the Windows platform

Add a stub EGL driver for Windows
Fix compiler issues in egl/main
Ensure Windows build produces libEGL.dll
Default EGL to enabled for Windows when building a Gallium driver

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

meson: Include EGL after gallium

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Add stw_* DLL exports for EGL support

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Swap buffers via pointer instead of HDC

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Add a stw_dev getter

Data imports need to be marked __declspec(dllimport), so
just export a function instead of data.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Use HWND instead of HDC as primary framebuffer handle

EGL's native window is an HWND, so this removes the need to
GetDC from the creation path there.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Add an explicit iPixelFormat for context creation

If it's 0, then it's looked up from the framebuffer for the specified HDC

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Un-inline helpers which use stw_own_mutex

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Add iPixelFormat to stw_pixelformat_info

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Split DrvReleaseContext to support unbind via pointer

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Make contexts current with framebuffers instead of HDCs

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Allow creating framebuffers that aren't in the global window list

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Make contexts current with pointer instead of DHGLRC

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Pass share context as pointer instead of DHGLRC

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

wgl: Create contexts and DHGLRCs separately

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Acked-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12727>

zink: stop setting nr_samples for null surfaces

this is a special value that shouldn't be set

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12875>

zink: don't copy inner surface refcount

this leaks surfaces. lots of surfaces.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12875>

intel/decode/gfx6: add support for gfx6 CC/VIEWPORT pointers.

These have 3 sub states encoded and valid fields on SNB.

Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12864>

radv/ci: add a list of expected failures for VanGogh

Empty means 100% passrate like Sienna Cichlid.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12872>

gitlab-ci: refactor timeout constants and tweak timeout values

* Refactor timeouts and retry attempts constants to variables in the top
  of the python script.

* Increase LAVA job timeout value from 1 minute to 5 minutes, since the
  timeout detection is just a heuristic based on the log silence in LAVA
  devices. If we keep 1 minute timeout, maybe we could cancel jobs that
  have tasks which may take too long to respond. Also, one minute
  timeout is prone to misdetect scenarios when some network errors or
  slowness may happen.

* Increase polling rate to check if the job has started from 1 check
  every 30 seconds to 1 check every 10 seconds. Since it was taking 30
  seconds in the worst case to start to get the log output from a LAVA
  job. It is important to note that some LAVA jobs take less than 2
  minutes to finish, so a 10 second wait would be more suitable in those
  cases.

Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12870>

r300: make global variables const (if possible)

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12860>

r300: fix usage of COVERED_PTR_MASKING_ENABLE for r500

http://developer.amd.com/wordpress/media/2013/10/
R5xx_Acceleration_v1.5.pdf

On page 281 you can see that register ZB:ZB_BW_CNTL
is using 19th bit for enabling this feature on r500.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12861>

radv: advertise EXT_shader_atomic_float2 with LLVM 14+

The intrinsics aren't exposed on earlier LLVM versions.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12716>

ac/llvm: implement nir_intrinsic_global_atomic_{fmin,fmax}

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12716>

ac/llvm: implement nir_intrinsic_shared_atomic_{fmin,fmax}

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12716>

ac/llvm: implement nir_intrinsic_ssbo_atomic_{fmin,fmax}

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12716>

ac/llvm: implement nir_intrinsic_image_deref_atomic_{fmin,fmax}

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12716>

freedreno: a2xx: Handle samplerExternalOES like sampler2D

There should be no difference in handling these two samplers,
handle GLSL_SAMPLER_DIM_EXTERNAL just like GLSL_SAMPLER_DIM_2D
to fix "unimplemented sampler 6" error in case someone tries to
use samplerExternalOES in shader program.

Signed-off-by: Marek Vasut <marex@denx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12817>

v3dv: start using Broadcom's device identifiers

Instead of creating our own based on the V3D version. CTS waivers
are registered using a combination of VendorID and DeviceID, so if
we want to reuse any wavers filed by Broadcom we want to use the
same identifiers. We are already using the Broadcom VendorId, so
let's start using the same deviceID as well.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12865>

radv/llvm: fix vertex input fetches with 16-bit floats

Not ideal but ac/llvm and RADV works with integers, so passing a
16-bit float type would break more than it helps.

Fixes a few CTS with 16-bit float IO.

Fixes: 3fb229e010f ("ac,radeonsi: load VS inputs at the call site of nir_intrinsic_load_input")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12835>

ci/v3d: add piglit flake

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12866>

radv/llvm: fix using Wave32

The Wave32 pass manager has been removed a while ago.

Fixes: 94a1f45e15a ("ac/llvm: set target features per function instead of per target machine")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12833>

radv: reduce SQTT traffic when instruction timing is disabled

Some tokens can be excluded without instruction timing. This reduces
RGP capture sizes significantly.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12853>

radv: Disable DCC on storage images that cannot support DCC image stores

Additional work is needed for storage images with DCC without DCC image stores to not be broken.

Fixes black screens in Doom Eternal.

Fixes: #5345
Signed-off-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12818>

nir: Drop the unused instr arg for src/dest copy functions.

Now that we don't use ralloc, we don't need this arg to get at the right
ralloc ctx.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir: Switch from ralloc to malloc for NIR instructions.

By replacing the 48-byte ralloc header with our exec_node gc_node (16
bytes), runtime of shader-db on my system across this series drops
-4.21738% +/- 1.47757% (n=5).

Inspired by discussion on #5034.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir/tests: Fix transmuting an SSA dest to be non-SSA

With the de-ralloc changes, having the register dest not have its .reg
properly initialized caused crashes.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir/lower_phis_to_scalar: Use nir_instr_free() to free instrs.

Preparation for de-rallocing instrs.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir: Add all allocated instructions to a GC list.

Right now we're using ralloc to GC our NIR instructions, but ralloc has
significant overhead for its recursive nature so it would be nice to use a
simpler mechanism for GCing instructions.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir: Consistently pass the instr to nir_src_copy().

The arg says it's supposed to be the instr, not the shader.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir: Consistently pass the shader to the shader arg of instr creation.

We were using the ralloc parent in some places, which should work out to
be the shader I think, but to de-ralloc the instrs we should just pass the
existing shader pointer in.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir/from_ssa: Use nir_instr_free() to free instrs instead of ralloc.

This code was being tricky with passing a mem_ctx instead of the shader,
then freeing the mem_ctx when the pass was done and all the parallel
copies had been removed from the shader.  Use the right type for instr
creation and do a bit of manual list management to prepare the way for
non-ralloc NIR instrs.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir: Pull the instr list free function out to a helper.

With the de-rallocing, we're going to have some more places that free a
list of instrs.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir: Add a nir_instr_free() to replace ralloc_free(instr).

This will gain another step shortly.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11776>

nir/lower_gs_intrinsics: Make nir_lower_gs_intrinsics be idempotent

Calling this lower pass twice in a row would cause spurious
set_vertex_and_primitive_count(0, undef) intrinsics after the proper
set_vertex_and_primitive_count intrinsic.  This pretty much turns any
geometry shader into garbage.

Fix this by treating nir_intrinsic_emit_vertex_with_counter and
nir_intrinsic_end_primitive_with_counter just like the non-_with_counter
versions.  If no blocks would need set_vertex_and_primitive_count
intrinsics added, exit the pass before doing any work.  This prevents
the need for DCE to do extra clean up later.

Since this pass is potentially called multiple times via multiple
invocations of a finalize_nir callback, it is (hypothetically?) possible
that control flow could be changed to add new blocks that need this
intrinsic.  The check implemented in this commit should be robust
against that possibility.

v2: Add a_block_needs_set_vertex_and_primitive_count.  Suggested by
Timur.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12802>

nir/lower_gs_intrinsics: Return progress if append_set_vertex_and_primitive_count makes progress

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Fixes: 542d40d698a ("nir: Add new GS intrinsics that maintain a count of emitted vertices.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12802>

radeonsi: enable NGG passthrough when LDS is used, document the real constraints

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: fix incorrect comments about VGT_SHADER_STAGES_EN

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: don't set edgeflags for TES and blit VS

they are disabled (TES) or have no effect (blit VS)

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: use a trick to extract and pack edgeflags using fewer instructions

This removes 4 instructions from the prim export packing.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: convert gfx10_emit_ge_pc_alloc to radeon_opt_set_uconfig_reg

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: eliminate redundant SPI_SHADER_PGM_RSRC3/4_GS register writes

They don't change much.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: put si_pm4_state at the beginning of si_shader

instead of allocating it separately. This removes pointer indirections.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: move setting one GS shader key field out of si_shader_selector_key

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: move setting most TCS shader key fields out of si_shader_selector_key

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: set prefer_mono outside of si_shader_selector_key

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: precompute more spi_map code

This replaces vs_output_param_offset by vs_output_ps_input_cntl,
which is easier to use.

For geometry shaders, vs_output_ps_input_cntl is stored in the GS si_shader
structure, not gs_copy_shader. This requires that gs_copy_shader compilation
is finished before the GS main shader part, so that GS can initialize
vs_output_ps_input_cntl using the compiled GS copy shader.

output_semantic_to_slot becomes unused, so it's removed.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: unroll loops in si_emit_spi_map using 33 C++ template instantiations

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: inline si_get_ps_input_cntl because it has only one use

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: simplify si_emit_spi_map for back-face colors

Just precompute what we need in si_shader_info.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: precompute num_interp for si_emit_spi_map

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: interleave si_shader_info::input_* in memory for faster emit_spi_map

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: restructure si_get_ps_input_cntl for future refactoring

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: force flat for PrimID early in si_nir_scan_shader

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: eliminate a not-found conditional for PrimID in si_get_ps_input_cntl

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: determine num_vbos_in_user_sgprs from template arguments in draw_vbo

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: deduplicate si_compiler_ctx_state initialization

to remove it from si_update_shaders

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: precompute si_vgt_stages_key for NGG in si_shader

to remove this overhead from si_update_shaders

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: move flat shading VRS enablement out of si_update_shaders

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: move DB_SHADER_CONTROL update for PS out of si_update_shaders

It only depends on the pixel shader CSO and alpha test.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: check flatshade and sprite_coord_enable for spi_map in bind_rs_state

it doesn't need to be in si_update_shaders

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: optimize scratch buffer size updates using C++ template arguments

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: make si_update_shaders a C++ template in si_state_draw.cpp

This reduces driver overhead.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: unset SI_PREFETCH_* only when we unbind pm4 shader states

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: move PS inputs_read computation out of si_get_vs_key_outputs

to reduce overhead of si_update_shaders

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: move si_shader_io_get_unique_index calls out of si_get_vs_key_outputs

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: rewrite inlinable uniform states for shader keys in si_context

directly update the shader keys in si_context

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: update the VS shader key in set & bind functions and remove memsets

This decreases overhead of si_update_shaders and overall driver overhead.

The VS shader key portion related to VS inputs is updated in set & bind
functions. Other fields related to outputs are still updated
in si_shader_selector_key.

Now that all modified fields are set to 0 when not needed, and remove
the memsets.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>

radeonsi: clean up and clear VS shader key fields related to outputs

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>