compiler: Add SYSTEM_VALUE_IS_INDEXED_DRAW and instrinsics

This VS system value contains if the draw command used to start the
rendering was an indexed draw command or a non-indexed one
(~0/0 respectively). Useful to calculate the gl_BaseVertex as:
(SYSTEM_VALUE_IS_INDEXED_DRAW & SYSTEM_VALUE_FIRST_VERTEX).

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

radv: enable out-of-order rasterization by default

As the implementation is conservative, we can now enable it
by default. It can be disabled with RADV_DEBUG=nooutoforder.

Don't expect much more than 1% of improvements, but the gain
seems consistent.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

radv: only disable out-of-order rast for perfect occlusion queries

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

i965: Drop unused gen5 sampler default color struct.

Trivial.

i965: Make brw_vs_outputs_written static.

Drop a prototype.  Trivial.

i965/tex_image: Avoid the ASTC LDR workaround on gen9lp

Both the internal documentation and the results of testing this in the
CI suggest that this is unnecessary. Add the fixes tag because this
reduces an internal benchmark's startup time by about 17 seconds
(reported by Eero).

Fixes: 710b1d2e665 "i965/tex_image: Flush certain subnormal ASTC channel values"
Tested-by: Eero Tamminen <eero.t.tamminen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

freedreno: Fix ir3_cmdline.c build.

Fixes: 6487e7a30c9e ("nir: move GL specific passes to src/compiler/glsl")
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>

anv: Allow lookup of vkEnumerateInstanceVersion without an instance

Fixes: cbab2d1da5edfe9df27a010adf8b1aa9dbee473b
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

anv: Don't advertise Float64 or Int64 on HW without 64-bit types

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

radv: compute the number of subpass attachments correctly

Only count color attachments twice if resolves are used, also
account for the depth stencil attachment if present.

Cc: 18.0 18.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

radv: set fmask_surf_index on fmask surfaces.

This is needed for gfx9 and later for all fmask surface index.

(Mentioned by Marek on irc)

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

gallium/i915: fix PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE typo

Fixes: fffe5e2d14f807c ("gallium: add initial support for conservative
rasterization")
Trivial.

nvc0: add conservative rasterization support

Subpixel precision bias, dilation and the post-snap mode are supported on
GM200 and newer. The pre-snap mode is supported for triangle primitives on
GP100.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

st/mesa: add support for nvidia conservative rasterization extensions

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium: add initial support for conservative rasterization

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: add support for nvidia conservative rasterization extensions

Although the specs are written against compatibility GL 4.3 and allows core
profile and GLES2+, it is exposed for GL 1.0+ and GLES1 and GLES2+.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

glsl/tests: add GLSL_TYPE_UINT8, GLSL_TYPE_INT8 cases to switch statements

To silence warnings about unhandled switch values.
Untested otherwise.

v2: move the INT/UINT8 cases after the INT/UINT16 cases, per Eric.

Reviewed-by: Eric Anholt <eric@anholt.net>

tgsi: use enums instead of unsigned in ureg code

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

nir: move GL specific passes to src/compiler/glsl

With this we should have no passes in src/compiler/nir with any
dependencies on headers from core GL Mesa.

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

radv/winsys: fix leaking resources from bo's imported by fd

A bo's ref_count was not being initialized when imported from an fd.
Therefore, we would fail to free the resource during VkFreeMemory().

This patch fixes applications like hifi VR in threaded mode, which
perform frequent imports/releases of IPC shared memory.

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
CC: 18.0 18.1 <mesa-stable@lists.freedesktop.org>

i965/tiled_memcpy: ytiled_to_linear a cache line at a time

Similar to the transformation applied to linear_to_ytiled, also align
each readback from the ytiled source to a cacheline (i.e. transfer a
whole cacheline from the source before moving on to the next column).
This will allow us to utilize movntqda (_mm_stream_si128) in a
subsequent patch to obtain near WB readback performance when accessing
the uncached ytiled memory, an order of magnitude improvement.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965: Record mipmap resolver for unmapping

When mapping a region of the mipmap_tree, record which complementary
method to use to unmap it afterwards. By doing so we can avoid
duplicating the decision tree used when mapping and thereby eliminate
trivial errors that can be introduced if the two if-chains become out of
sync.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Scott D Phillips <scott.d.phillips@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Move unmap_depthstencil before map_depthstencil

Reorder code to avoid a forward declaration in the next patch.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Move unmap_etc before map_etc

Reorder code to avoid a forward declaration in the next patch.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Move unmap_s8 before map_s8

Reorder code to avoid a forward declaration in the next patch.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Move unmap_movntdqa before map_movntdqa

Reorder code to avoid a forward declaration in the next patch.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Move unmap_blit before map_blit

Reorder code to avoid a forward declaration in the next patch.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Move unmap_gtt before map_gtt

Reorder code to avoid a forward declaration in the next patch.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

ac/nir: expand 64-bit vec3 loads to fix shuffling.

If loading 64-bit vec3 values, a 4 component load would be followed
by a 2 component load and the resulting shuffle would fail as it
requires 2 4 components. This just expands the second results
vector out to 4 components.

This fixes 100 CTS tests:
dEQP-VK.spirv_assembly.type.vec3.*64*

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

i965: Don't stomp initial kflags for program cache.

We want to flag EXEC_OBJECT_CAPTURE, but we ought to preserve any
existing kflags.  Today, there are none (as the program cache doesn't
support 48-bit addressing), but once we start using softpin, we'll
need to preserve EXEC_OBJECT_PINNED.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

i965: Let batchbuffers be placed anywhere in the 48-bit address space.

We were trying to mark batch buffers with EXEC_OBJECT_CAPTURE, and
accidentally stomped EXEC_OBJECT_SUPPORTS_48B_ADDRESS in the process.

There's no reason to restrict batch buffers to the lower 4GB.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

intel: fix check for 48b ppgtt support

The previous logic of the supports_48b_addresses wasn't actually
checking if i915.ko was running with full_48bit_ppgtt. The ENOENT
it was checking for was actually coming from the invalid context
id provided in the test execbuffer.  There is no path in the
kernel driver where the presence of
EXEC_OBJECT_SUPPORTS_48B_ADDRESS leads to an error.

Instead, check the default context's GTT_SIZE param for a value
greater than 4 GiB

v2 (Ken): Fix in i965 as well.
v3 Check GTT_SIZE instead of HAS_ALIASING_PPGTT (Chris Wilson)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

st/omx/enc: fix blit setup for YUV LoadImage

The blit here involves scaling since it's copying from I8 format to R8G8 format.
Half of source will be filtered out with PIPE_TEX_FILTER_NEAREST instruction, it
looks that GPU always uses the second half as source. Currently we use "1" as
the start point of x for R, then causing 1 source pixel of U component shift to
right. So "-1" should be the start point for U component.

Cc: 18.0 18.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

autotools, meson: bump up required VA version

Due using a new VP9 config we use, required VA API 0.39

Fixes: 413c5ca3727 ("travis: update libva required version")
CC: 18.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

docs: update calendar, add news and link release notes to 18.0.2

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>

docs: add sha256 checksums for 18.0.2

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
(cherry picked from commit b3eed3ad03fd1eb61474cd0a8a173ad40fb8a876)

docs: add release notes for 18.0.2

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
(cherry picked from commit d38da7bd2d4387635fac8bc7f45e64f50dc43c43)

radeonsi: increase the number of compiler threads depending on the CPU

The compiler queue was limited to 3 threads, so shader-db running
on a 16-thread CPU would have a bottleneck on the 3-thread queue.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: avoid a crash in gallivm_dispose_target_library_info

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: move data_layout into si_compiler

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: move passmgr into si_compiler

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: move target_library_info into si_compiler

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: use si_compiler::triple in si_llvm_optimize_module

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: add triple into si_compiler

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: add struct si_compiler containing LLVMTargetMachineRef

It will contain more variables.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Benedikt Schemmer <ben at besd.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: rename r600_texture::resource to buffer

r600_resource could be renamed to si_buffer.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: use r600_resource() typecast helper

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: remove unused atom parameter from si_atom::emit

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: inline 2 trivial state structures

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: remove function si_init_atom

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: remove si_atom::id

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: don't use an indirect table for state atoms

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: rename r600_atom -> si_atom

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: remove r600_pipe_common.h

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi/gfx9: workaround for INTERP with indirect indexing

and clean up the conditions.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Cc: 18.0 18.1 <mesa-stable@lists.freedesktop.org>

radeonsi: rewrite DCC format compatibility checking code

It might be better to use a slow compressed clear when clearing to 1.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: implement DCC fast clear swizzle constraints more accurately

Reduce swizzle constraints to the ALPHA_IS_ON_MSB constraint and the clear
value of 1.

This significantly changes the DCC fast clear code, and fixes fast clear
for RGB formats without alpha.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: rename variables and document stuff around DCC fast clear

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: fully enable 2x DCC MSAA for array and non-array textures

The clear code is exactly the same as for 1 sample buffers -
just clear the whole thing.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: enable fast color clear for level 0 of mipmapped textures on <= VI

GFX9 is more complicated and needs a compute shader that we should just
copy from amdvlk.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

ac/surface: handle DCC subresource fast clear restriction on VI

v2: require the previous level to be clearable for determining whether
    the last unaligned level is clearable

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

swr/rast: No need to export GetSimdValidIndicesGfx

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Small editorial changes

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Use new processor detection mechanism

Use specific avx512 selection mechanism based on avx512er bit instead of
getHostCPUName().  LLVM 6.0.0 has a bug that reports wrong string for KNL
(fixed in 6.0.1).

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Output rasterizer dir to console since it's process specific

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Add TranslateGfxAddress for shader

Also add GFX_MEM_CLIENT_SHADER

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: jit PRINT improvements.

Sign-extend integer types to 32bit when specifying "%d" and add new %u
which zero-extends to 32bit. Improves  printing of sub 32bit integer types
(i1 specifically).

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Fix regressions.

Bump jit cache revision number to force recompile.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Cleanup old cruft.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Package events.proto with core output

However only if the file exists in DEBUG_OUTPUT_DIR. The expectation is
that AR rasterizerLauncher will start placing it there when launching
a workload (which is in a subsequent checkin)

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Fix init in EventHandlerWorkerStats

Make sure we initialize variables.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Fix return type of VCVTPS2PH.

expecting <8xi16> return.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: WIP Translation handling

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Use different handing for stream masks

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Silence warnings

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Add support for TexelMask evaluation

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Internal core change

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Fix x86 lowering 64-bit float handling

- 64-bit cvt-to-float needs to be explicitly handled
- gathers need the right parameter types to work with doubles

Fixes draw-vertices piglit tests

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Add some SIMD_T utility functors

VecEqual and VecHash

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Fix wrong type allocation

ALLOCA pointer elements, not pointers.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr: touch generated files to update timestamp

previous change in generators necessitates this change

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: Fix byte offset for non-indexed draws

for the case when USE_SIMD16_SHADERS == FALSE

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

util/u_queue: fix a deadlock in util_queue_finish

Cc: 18.0 18.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

meson: fix race condition revealed by using 0.44

Previously there was a special target that blocked for the generation of
anv_entrypoints.h, with meson 0.44 we don't need this, we can use a new
language feature instead. The problem is that previously that blocking
target would hide a race condition for the generation of another header,
anv_extensions.h. Now the build sometimes fails when anv_extensions.h is
not generated in time.

v2: - clarify the race condition in the commit message (Emil)

CC: Mark Janes <mark.a.janes@intel.com>
Fixes: 92550d9b16d2b295bdac087f31b1fd6d0f808e02
       ("meson: remove workaround for custom target creating .h and .c files")
Signed-off-by: Dylan Baker <dylan.c.baker@intel.com>
Reviewed-by: Scott D Phillips <scott.d.phillips@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

bin: force git show to use default pretty setting

I have pretty default to short, which breaks this script.

v2: - Fix both places that don't define a --pretty (Emil)

cc: Juan A. Suarez <jasuarez@igalia.com>
Signed-off-by: Dylan Baker <dylan.c.baker@intel.com>
Reviewed-by: Andres Gomez <agomez@igalia.com> (v1)
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

mesa: add TBO support for GL_EXT_texture_norm16

Earlier plumbing missed interaction with texture buffer objects.

Fixes: 7f467d4f73 "mesa: GL_EXT_texture_norm16 extension plumbing"
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

ac: fix texture query LOD for 1D textures on GFX9

1D textures are allocated as 2D which means we only need
one coordinate for texture query LOD.

Fixes: 625dcbbc456 ("amd/common: pass address components individually to
ac_build_image_intrinsic")
Cc: 18.1 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

etnaviv: remove not needed includes

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

etnaviv: remove redundant include

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

glsl: replace some asserts with unreachable when processing the ast

Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa: drop the buffer mode param from the DrawBuffer driver function

No drivers used it.

Reviewed-by: Brian Paul <brianp@vmware.com>

anv/icl: Enable Vulkan on Ice Lake

This patch enables the Vulkan driver on Ice Lake h/w
with added warning about preliminary support.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv: enable VK_EXT_shader_viewport_index_layer

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/allocator: Don't shrink either end of the block pool

Previously, we only tried to ensure that we didn't shrink either end
below what was already handed out.  However, due to the way we handle
relocations with block pools, we can't shrink the back end at all.  It's
probably best to not shrink in either direction.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105374
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106147
Tested-by: Eero Tamminen <eero.t.tamminen@intel.com>
Reviewed-by: Scott D Phillips <scott.d.phillips@intel.com>
Cc: mesa-stable@lists.freedesktop.org

broadcom/vc5: Add support for centroid varyings.

It would be nice to share the flags packet emit logic with flat shade
flags, but I couldn't come up with a good way while still using our pack
macros.  We need to refactor this to shader record setup at compile time,
anyway.

Fixes ext_framebuffer_multisample-interpolation * centroid-*

broadcom/vc5: Add an assert about GFXH-1559.

Our TF outputs always start at 6 or 7 currently, so we don't hit the
broken 8 case.  Let's make sure that doesn't change somehow.

broadcom/vc5: Add validation that we don't violate GFXH-1633 requirements.

We don't use ldunifa yet, but we will eventually for UBOs.

broadcom/vc5: Add validation that we don't violate GFXH-1625 requirements.

We don't use TMUWT yet, but we will once we do SSBOs.

broadcom/vc5: Implement GFXH-1742 workaround (emit 2 dummy stores on 4.x).

This should fix help with intermittent GPU hangs in tests switching
formats while rendering small frames.  Unfortunately, it didn't help with
the tests I'm having troubles with.

broadcom/vc5: Add QPU validation for register writes after thrend.

The next shader gets to start writing the register file during these
slots, so make sure we don't stomp over them.

The only case of hitting this that I could imagine would be dead writes.