pipe-loader: use radeonsi for MM if amdgpu dri is used

The amdgpu dri is used for the closed source AMD driver. Since this driver
does not implement multimedia, we fall back to radeonsi in mesa to do
multimedia. This corrects the dri driver name for when it is set to amdgpu.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com> (v1)
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

egl: drop incorrect pkg-config file for glvnd

With b01524fff05eef66e8cd ("meson: don't build libGLES*.so with GLVND")
we dropped the incorrect pkg-config files for GLES*.

Since then, the glvnd issue of its missing files has become painfully
apparent, since it break the build for everyone using glvnd.

NVIDIA has had a fix for a few years now, but has yet to accept it:
https://github.com/NVIDIA/libglvnd/pull/86

Since the breakage is already there, let's clean up everything on our side
while we wait for NVIDIA to accept the fix.

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

docs: simplify `Fixes:` git command

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>

mesa/tests: add missing dep_thread

Fixes: f8c27c277585141f2d27 ("state_tracker: Move the format test out to be an actual unit test.")
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Vinson Lee <vlee@freedesktop.org>

util: drop strncat(), strcmp(), strncmp(), snprintf() & vsnprintf() MSVC fallbacks

It would seem MSVC>=2015 is now C99-compliant wrt these functions:
strncat:   https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/strncat-strncat-l-wcsncat-wcsncat-l-mbsncat-mbsncat-l?view=vs-2017
strcmp:    https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/strcmp-wcscmp-mbscmp?view=vs-2017
strncmp:   https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/strncmp-wcsncmp-mbsncmp-mbsncmp-l?view=vs-2017
snprintf:  https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/snprintf-snprintf-snprintf-l-snwprintf-snwprintf-l?view=vs-2017
vsnprintf: https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/vsnprintf-vsnprintf-vsnprintf-l-vsnwprintf-vsnwprintf-l?view=vs-2017

Suggested-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for vsnprintf()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for snprintf()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for vasprintf()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for sprintf()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for strcmp()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for strcasecmp()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for strncmp()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for strncat()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for strdup()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: use standard name for strchrnul()

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: drop unused vsprintf() wrapper

Suggested-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: drop unused strchr() wrapper

Suggested-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

util: drop unused strstr() wrapper

Suggested-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

nir: Only rematerialize comparisons with all SSA sources

Otherwise, you may end up moving a register read and that could result
in an incorrect shader.  This commit fixes a rendering issue in Elite:
Dangerous.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111152
Fixes: 3ee2e84c60 "nir: Rematerialize compare instructions"
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

spirv: Fix order of barriers in SpvOpControlBarrier

Semantically, the memory barrier has to come first to wait
for the completion of pending memory requests.
Afterwards, the workgroups can be synchronized.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

nir: use a switch when printing intrinsic indices

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

nir/algebraic: mark a few comparison simplifications as precise

No vkpipeline-db changes found.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reveiewed-by: Alyssa Rosenzweig alyssa.rosenzweig@collabora.com
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/algebraic: optimize contradictory iand operands

Some of these were found in a few GTAV, Rise of the Tomb Raider and
Shadow of the Tomb Raider shaders.

Results from vkpipeline-db run with ACO:
Totals from affected shaders:
SGPRS: 376 -> 376 (0.00 %)
VGPRS: 220 -> 220 (0.00 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 13492 -> 11560 (-14.32 %) bytes
LDS: 6 -> 6 (0.00 %) blocks
Max Waves: 69 -> 69 (0.00 %)
Wait states: 0 -> 0 (0.00 %)

v2: use False instead of 0

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reveiewed-by: Alyssa Rosenzweig alyssa.rosenzweig@collabora.com
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

lima/ppir: handle all node types in ppir_node_replace_child

ppir_node_replace_child is used by the const lowering routine in ppir.
All types need to be handled here, otherwise the src node is not updated
properly when one of the lowered nodes is a const, which results in, for
example, regalloc not assigning registers correctly.

Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>

lima/ppir: branch regalloc fixes

The branch instruction has sources which must be handled in src handling
paths so that regalloc assigns registers to them properly.

Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>

main: Destroy static hash table

format_array_format_table has a static lifetime - it will be destroyed
by an atexit handler.

Signed-off-by: Yevhenii Kolesnikov <yevhenii.kolesnikov@globallogic.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

radv: reset the window scissor with no clear state.

If we don't have clear state (which gfx10 doesn't currently)
we will fix to reset the scissor. AMDVLK will leave it set
to something else.

Marek also has this fix for radeonsi pending.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

radv: fix crash in shader tracing.

Enabling tracing, and then having a vmfault, can leads to a segfault
before we print out the traces, as if a meta shader is executing
and we don't have the NIR for it.

Just pass the stage and give back a default.

Fixes: 9b9ccee4d64 ("radv: take LDS into account for compute shader occupancy stats")
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

iris: change last_vue_stage() to look at uncompiled shaders

This allows us to find the last vue stage before we have compiled
the shaders.

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

nir/lower_clip: add support for geometry shaders

This will be used to enabled compat profile support for geometry
shaders.

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

nir/lower_clip: add lower_clip_outputs() helper

This will be reused in the following patch to add support for clip
vertex lowering in geometry shaders.

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

nir/lower_clip: add create_clipdist_vars() helper

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

nir/lower_clip: add a find_clipvertex_and_position_outputs() helper

This will allow code sharing in a following patch that adds support
for lowering in geometry shaders. It also allows us to exit early
if there is no lowering to do which allows a small code tidy up.

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

panfrost: Set rt_count

This doesn't quite work yet, but it illustrates how MRT is implemented
in the MFBD: rt_count is set appropriately based on the number of render
targets, while additional render target descriptors are appended on with
an index variable in them (not quite decoded since there's some aspects
we don't understand there, but conceptually this should be right).

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Trace invisible BOs

Helps make the decode a little more readable (names instead of
addresses).

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost/decode: Preserve empty tiler heap symmetry

If tiler_heap_end == tiler_heap_start, ensure it's printed the same
rather than one erroring out as hex.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Zero polygon list body size for clears

There's no polygons, so you can't have any size to the polygon list,
although there is a minimal header.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost/mfbd: Unify depth-only with masked FBO path

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Simplify set_framebuffer_state

Most of the ad hoc logic is already in Gallium.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Check for NULL surface in places

Fixes a bunch of NULL dereferences, although it does cause GPU faults of
course.

This is caused by color buffers masked out in MRT, which we'll
eventually have to solve the right way... one thing at a time.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Expose 4 render targets

Hidden behind deqp flag as usual.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Shrink tiler heap

128MB is excessive and 16MB is still plenty. Saves 112MB/context on
kernels without growable/heap support.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

nir/large_constants: De-duplicate constants

If a function has a constant and is called more than once, after
inlining we may end up with different variables representing the same
constant.  This commit look into the data and de-duplicate them.

The first pass now will collect the constant data in a per variable
buffer, then de-duplication happens (by sorting then linear walk), and
the second pass will use the data in var->data.location.

One side-effect of the current implementation is that constants will
be reordered.  If this turns out to be a problem is something that can
be fixed.

An alternative strategy considered was to perform this in a
per-function basis and then merge the results, the problem is that we
would have to fix up the offsets during the merge.  Given the data we
have, the current patch is good enough.

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

nir/large_constants: Use ralloc for var_infos

This will be used later on to allocate constant data for each
variable (and then deduplicate).  Also drop initializing found_read,
as it is already implicitly false in the literal.

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

freedreno: Convert nir_lower_tg4_to_tex to the NIR lowering helper.

Cuts a bunch of boilerplate.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>

freedreno: Convert load_barycentric_at_sample to the NIR lowering helper.

Cuts out a ton of boilerplate.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>

freedreno: Convert load_barycentric_at_offset to the NIR lowering helper.

Cuts out a ton of boilerplate.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>

v3d: Use nir_shader_lower_instructions() for txf_ms lowering.

Cuts out a bunch of boilerplate.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

nir: Allow internal changes to the instr in nir_shader_lower_instructions().

v3d's NIR txf_ms lowering wants to swizzle around the input coordinates in
NIR, but doesn't generate a new txf_ms instructions as replacement.  It's
pretty easy to allow that in nir_shader_lower_instructions, and it may be
common in lowering passes.

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

vc4: Convert vc4_nir_lower_txf_ms to nir_shader_lower_instructions().

Cuts out a bunch of boilerplate.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

v3d: Fix assertion failures in debug builds.

nir_lower_io leaves around deref_var instructions after lowering away
deref intrinsics.  This ends up breaking validation after v3d_nir_lower_io
removes variables not actually being stored by the shader's
store_output()s.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

panfrost: Handle Z24 textures

Just use the Z32 code.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost/ci: Update expectations

We just fixed some stencil tests.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Make scissor test more robust

See v3d implementation.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Use correct NO_DITHER field on MFBD

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Implement Z32F(_S8) support

Z32F uses a dediacted float path. Z32F_S8 uses separate stencil planes
in the hardware, lowered via u_transfer_helper.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost/decode: Don't disassemble NULL shaders

It is legal to load a shader from a NULL address, particularly when the
TILER job is used strictly for effects on the Z/S buffer with 0x0 color
mask. Don't crash the decoder in this case.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Copy stencil front to back if back disabled

When backside stenciling is disabled, backfacing primitives just do the
same thing as frontfacing primitives.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

swr/rast: Refactor memory API between rasterizer core and swr

This commit cleans up API between the core of the rasterizer and swr.
Some formatting changes are also done.

Reviewed-by: Alok Hota <alok.hota@intel.com>

lima/ppir: Add gl_PointCoord handling

Treat gl_PointCoord as a system value and
add the necessary bits for correct codegen.

Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

gallium: Add PIPE_CAP_TGSI_FS_POINT_IS_SYSVAL

This adds an option to treat gl_PointCoord as a system value.

Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

nir/tgsi: Extend tgsi_to_nir.c to support gl_PointCoord as a system value.

Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Eric Anholt <eric@anholt.net>

nir: Add gl_PointCoord system value

gl_PointCoord handling needs some special bits set in lima/ppir code
generation. Treating gl_PointCoord as a system value makes it easier
to distinguish from a regular varying.

Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Optionally declare gl_PointCoord as a system value

Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

lima/gp: Fix problem with complex moves

When writing the scheduler, we forgot that you can't read the complex
unit in certain sources because it gets overwritten to 0 or 1. Fixing
this turned out to be possible without giving up and reducing
GPIR_VALUE_REG_NUM to 10, although it was difficult in a way I didn't
expect. There can be at most 4 next-max nodes that can't have moves
scheduled in the complex slot, so it actually isn't a problem for
getting the number of next-max nodes at 5 or lower. However, it is a
problem for stores. If a given node is a next-max node whose move cannot
go in the complex slot *and* is used by a store that we decide to
schedule, we have to reserve one of the non-complex slots for a move
instead of all the slots, or we can wind up in a situation where only
the complex slot is free and we fail the move. This means that we have
to add another term to the reservation logic, for stores whose children
cannot be in the complex slot.

Acked-by: Qiang Yu <yuq825@gmail.com>

lima/gpir: Rework the scheduler

Now, we do scheduling at the same time as value register allocation. The
ready list now acts similarly to the array of registers in
value_regalloc, keeping us from running out of slots. Before this, the
value register allocator wasn't aware of the scheduling constraints of
the actual machine, which meant that it sometimes chose the wrong false
dependencies to insert. Now, we assign value registers at the same time
as we actually schedule instructions, making its choices reflect reality
much better. It was also conservative in some cases where the new scheme
doesn't have to be. For example, in something like:

1 = ld_att
2 = ld_uni
3 = add 1, 2

It's possible that one of 1 and 2 can't be scheduled in the same
instruction as 3, meaning that a move needs to be inserted, so the value
register allocator needs to assume that this sequence requires two
registers. But when actually scheduling, we could discover that 1, 2,
and 3 can all be scheduled together, so that they only require one
register. The new scheduler speculatively inserts the instruction under
consideration, as well as all of its child load instructions, and then
counts the number of live value registers after all is said and done.
This lets us be more aggressive with scheduling when we're close to the
limit.

With the new scheduler, the kmscube vertex shader is now scheduled in 40
instructions, versus 66 before.

Acked-by: Qiang Yu <yuq825@gmail.com>

lima/gp: Mark more add-only nodes as maybe-two-slot

Reviewed-by: Qiang Yu <yuq825@gmail.com>

lima/gpir: Fix some bugs in instruction handling

Reviewed-by: Qiang Yu <yuq825@gmail.com>

lima: Reintroduce the standalone compiler

I used this to test things without needing to have a device handy.

Acked-by: Qiang Yu <yuq825@gmail.com>

nir/lower_viewport: Check variable mode first

The location is unused for shader_temp and function_temp variables, and
due to the way we nir_lower_io_to_temproraries demotes shader_out
variables to shader_temp variables, it happened to equal
VARYING_SLOT_POS for the gl_Position temporary, which made this pass
fail with the offline compiler due to this coming before vars_to_ssa.

Reviewed-by: Qiang Yu <yuq825@gmail.com>

radv/gfx10: set BREAK_WAVE_AT_EOI if TES or GS enable the primitive ID

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

radv/gfx10: move emitting VGT_PRIMITIVEID_EN into the NGG path

And do not emit VGT_GS_MODE which is unnecessary on GFX10.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

radv/gfx10: do not always execute a barrier before the second shader

With NGG, empty waves may still be required to export data.

This fixes dEQP-VK.ycbcr.format.*_unorm.geometry_*.

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

radv: fix VGT_GS_MODE if VS uses the primitive ID

Found by inspection.

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

v3d: emit correct lowering for logic operations with MSAA render targets

v2:
 - Drop the writemask from the per-sample color intrinsic (Eric)

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

v3d: handle nir_intrinsic_store_tlb_sample_color_v3d

v2:
 - Move handling of output intrinsics to ntq_emit_intrinsic() (Eric).

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

nir: add a V3D-specific intrinsic for per-sample color writes

For per-sample color writes we need the output intrinsic to pack the
sample index, which is not provided with regular store_output intrinsics
unless we figured out a way to encode it into the base or the offset.

v2:
 - Drop the writemask (Eric)

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

v3d: implement per-sample tlb color writes

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

v3d: refactor the tlb color write code

We want to split the tlb specifier setup from the color writes, because when
we implement per-sample color writes we want to do the latter for all the
samples, but the former only once.

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

v3d: move tlb color write emission to a helper function

We will soon be adding per-sample color writes which means additional
complexity and more indentation (we will need another loop to emit
the writes for each individual sample), so this will help keeping
things simple and a bit more readable.

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

v3d: implement per-sample tlb color reads

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

anv: fix format mapping for depth/stencil formats

anv_format is supposed to have a pointer back to the associated
VkFormat, we were missed this for depth/stencil formats.

This doesn't fix anything afaict, but will be needed for future
changes.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 465de47bad70 ("anv: associate vulkan formats with aspects")
Acked-by: Jason Ekstrand <jason@jlekstrand.net>

radv: put back VGT_FLUSH at ring init on gfx10

I can find no evidence that removing this is a good idea.

Fixes: 9b116173b6a ("radv: do not emit VGT_FLUSH on GFX10")
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

softpipe: Clamp border colors when needed

unorm and snorm require that the border color values are clamped, so when
picking the sampler view copy/clamp the border color from the sampler and
use these adjusted values.

Fixes:

  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_compressed_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_snorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_srgb_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.linear_unorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_compressed_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_snorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_srgb_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_unorm_color
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_unorm_depth
  dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_unorm_depth_uint_stencil_sample_depth

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

softpipe: set a lower minimum clamp value for texture coordinate border clamp

The value of -0.5f is not small enough to produce negative coordinates,
so lower the minimum clamp value to -1.0f. This fixes a number of tests
from
   dEQP-GLES31.functional.texture.border_clamp.*

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

softpipe: Correct repeat-mirror evaluation

when mirroring the texture corrdinates the indices must be mirrored as
well and the half pixel shift must be applied in reverse.

Fixes a number of tests from:
  dEQP-GLES31.functional.texture.gather.offset.*
  dEQP-GLES31.functional.texture.gather.offsets.*

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

softpipe: Also mark textures as dirty when updating the framebuffer state

At this point all the draw caches are flushed to the old attached textures,
so the read caches of these textures will need to be updated too.

Fixes:
   dEQP-GLES3.functional.fbo.color.repeated_clear.sample.tex2d.*

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

etnaviv: set DITHER_MODE

This fixes a rendering glitch observed in SDL testscale test, where alpha
blending samples with value (1.0, 1.0, 1.0, 0.0) whitens the target instead
of having no effect.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

etnaviv: update headers from rnndb

Update to etna_viv commit a16a418.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

etnaviv: fix blend color on newer GPUs

Newer GPUs use the half float ALPHA_COLOR_EXT register.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

etnaviv: fix alpha blending cases

We need to check rgb_func/alpha_func when determining if blend or separate
alpha is required.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

etnaviv: fix polygon offset

Dividing the fui result by 65535 is obviously wrong, and from testing, on
GC7000L at least there is no division by 65535.

Fixes dEQP-GLES2.functional.polygon_offset.fixed16_displacement_with_units

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

radv: dont store disasm string unless keep_shader_info flag set

This fixes the memory use regression from bug 111107.

Fixes: 726a31df705 ("radv: Add the concept of radv shader binaries.")

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111107

radv/gfx10: set the pgm rsrc3/4 regs using index sh reg set

This is ported from AMDVLK, it's probably not requires unless
we want to use "real time queues", but it might be nice to just have
in place.

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

radv: use correct register setter for ngg hw addr

this shouldn't matter, but it's good to be correct.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

freedreno/a6xx: Drop the WFI in the program update stateobj.

Rob Clark thinks this was likely a workaround for our const buffer
update bugs, and now that it's passing tests, we should be able to
drop it.

renderdoc-traces results:

traces/android/clashofclans.rdc:  +6.1% +/-   1.1%
traces/android/candycrush.rdc:    +5.2% +/-   1.6%

Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno/a6xx: Drop the WFI in constant uploads.

Now that the bin vs render constlen is fixed, we can skip these waits.

Improves webgl aquarium performance at 10k fish from 27fps to 33.
Some highlights from renderdoc-traces:

traces/android/minecraft.rdc:             +17.1% +/-   3.4%
traces/glmark2/ideas-speed=duration.rdc:  +11.6% +/-   2.4%
traces/android/candycrush.rdc:            +5.4%  +/-   1.1%
traces/android/clashofclans.rdc:          +4.4%  +/-   1.3%

Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno: Assert that we don't exceed constlen.

We actually could go up to vs->constlen in the binning shader on a6xx,
but for sanity let's make sure that we're always under constlen.  This
would have caught the bug fixed in 572c76fd8826 ("freedreno: Clamp UBO
uploads to the constlen decided by the shader.")

Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno: Fix more constlen overflows.

Fixes constlen overflow in
dEQP-GLES31.functional.shaders.builtin_var.compute.num_work_groups and
dEQP-GLES31.functional.image_load_store.buffer.image_size.readonly_32
and probably others.

Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno: Drop stale comment about skipping uploads.

We already skip the upload if it's unused, due to the constlen >
offset check.

Reviewed-by: Rob Clark <robdclark@gmail.com>