intel/compiler: Drop opt_sampler_eot()

Gen9 and Cherryview have the ability to mark texture instructions with
the End-of-thread bit under some conditions, which allows the texture
result to be written to the render target directly, rather than
returning to the EU.

In order to handle overlapping primitives correctly, we have to use the
'sendc' instruction which stalls until other threads potentially writing
to the same locations in the render target are retired. Unfortunately,
this stall happens before the texture is sampled (rather than in
parallel with stall), so for some literal edge cases (like the diagonal
edge between two triangles forming a rectangle) there can be a
performance penalty. As a result, it's probably not a good idea to use
this optimization in general.

I had planned to leave it enabled only for BLORP, where we use rectangle
primitives and are typically clearing/blitting an entire render target
without any overlapping primitives, but I noticed that the optimization
wasn't applied in some normal cases anyway. For example, in the piglit
test tests/shaders/glsl-fs-texture2d-bias.shader_test it is applied to
one BLORP-blit shader but not another due to some kind of mishandling of
register types (the destination register type of the texture operation
is UD while the color source of the render target write is F).

Additionally the instruction scheduler assumed that the combined texture
and render target write operation took 0 cycles, leading to cycle
estimates that are wildly inaccurate. Since the optimization was not
implemented for SIMD32 and our decision whether to use the SIMD32
program is made by comparing the estimated performance with that of the
SIMD16 shader, we wrongly threw out a bunch of SIMD32 programs that are
likely profitable.

   total cycles in shared programs: 472807891 -> 473784245 (0.21%)
   cycles in affected programs: 108277 -> 1084631 (901.72%)
   helped: 0
   HURT: 1290

   total sends in shared programs: 998955 -> 1000245 (0.13%)
   sends in affected programs: 1400 -> 2690 (92.14%)
   helped: 0
   HURT: 1290

   LOST:   0
   GAINED: 33

This patch shows no performance changes in Intel's Mesa performance CI.

Given the problems, the lack of evidence that the pass improves
performance, and the fact that the hardware feature was removed from
subsequent GPU generations, I think that the pass is not valuable and
should be removed.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5412>

ci: Disable some flaky tests on turnip.

These have appeared more than once in the flake reporting channel, and a
couple of them have spuriously failed marge-bot merges.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5429>

ci: Fix weird filesystem globs appearing in failed test .qpa files.

When you get a filure and go looking in the results, you'll find weird stuff like this in the XML:

Reference images fill undefined pixels with 3x3 grid pattern.
Attachment 0 (p' = p bin boot builds
dEQP-VK.renderpass.suballocation.attachment_allocation.grow_shrink.89.qpa
deqp dev etc home init install lib media mnt proc results root run sbin
set-job-env-vars.sh sys tmp usr var (1, 1, 1, 1) + (-1, -1, -1, 1))

because we were not quoting the line and 'p *' was getting expanded.

Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5435>

docs: update calendar, add news item, and link releases notes for 20.0.8

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5444>

docs: Add sha256sums for 20.0.8

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5444>

docs: Add release notes for 20.0.8

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5444>

pan/mdg: Prefer type over regmode for schedule constraints

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5443>

pan/mdg: Analyze types for 64-bitness in RA

Instead of reg_mode.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5443>

pan/mdg: Explicitly type 64-bit uniform moves

Instead of relying on reg_mode.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5443>

turnip: add emit renderpass cache flushes for sysmem 3D CmdClearAttachments

This clear path behaves like a draw, it needs the same flush as tu_draw.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5426>

turnip: add layered 3D path clear for CmdClearAttachments

This fixes cases where the 3D path is used with layered rendering.

Fixes dEQP-VK.renderpass.suballocation.multisample_resolve.layers* failures

Note the blob's 3D fallback path behaves differently, and uses the
framebuffer information to clear each layer individually (changing the MRT
state each time). But that's not possible in all cases, and the blob fails
to clear properly in dEQP-VK.geometry.layered.*.secondary_cmd_buffer cases.
So this clear path is not based on the blob's behavior.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5426>

turnip: share code between 3D blit/clear path and tu_pipeline

Instead of filling out registers manually, fill out ir3 structs and re-use
code from tu_pipeline.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5426>

turnip: pipeline program state refactor

This refactor simplifies things a bit, and will make it easier to share
some logic with tu_clear_blit (see next patches).

This changes the order in which some things are emitted, and emits less
for disabled shader stages. There's also as extra write to SP_GS_PRIM_SIZE
that is removed.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5426>

panfrost: Demote mediump varyings to fp16

Likewise lowp.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Override varying format to minimal precision

Spec allows this!

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Use shader_info harder

We already have this metadata..

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Only store varying formats

This reduces linking complexity.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Allow R/RG/RGB varyings

This can be a bandwidth savings.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Remove unused routines

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Use new varying linking

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Add high-level varying emit

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Add helper to determine if we are capturing

That is, is the varying setup for xfb *and* is there a buffer for it?

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Emit xfb records

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Emit special varyings

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Emit unlinked varyings

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Determine varying buffer presence

Essentially the same logic as before, but the assumptions are much more
explicit.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Introduce bitfields for tracking varyings

Rather than having all sorts of random state flyng about with varying
emission, we can use a simple present mask and general stride to encode
everything we need for non-XFB cases, and layer XFB on top easily
enough.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Add panfrost_streamout_offset helper

Calculates the bias required for an xfb record in the src_offset field
to account for truncating the address to force alignment.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

panfrost: Calculate varying size by format

Will enable <16-byte varyings.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

pan/mdg: Avoid fusing ld_vary_16 with non-zero component

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5423>

aco: fix shared subdword loads

Shared subdword loads don't need byte alignment as they are split
into multiple loads if necessary.

Fixes: 5cde4989d3c8c25b0ba2a11ec450625e30092b16 ('aco: remove unnecessary split- and create_vector instructions for subdword loads')
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5441>

radv: enable radv_enable_mrt_output_nan_fixup for RAGE 2

To fix game artifacts. It's always sad to have to fix game bugs
inside drivers ...

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/5359>

radv/llvm: implement radv_enable_mrt_output_nan_fixup workaround

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/5359>

aco: implement radv_enable_mrt_output_nan_fixup workaround

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5359>

radv: add new drirc option radv_enable_mrt_output_nan_fixup

To replace NaN from FS with zeros to fix game bugs.

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/5359>

glsl: fix incorrect optimisation in opt_constant_variable()

When handling function inputs the optimisation pass incorrectly
assumes the inputs are undefined. Here we simply change things to
assume inputs have always been assigned a value. Any further
optimisations will be taken care of once function inlining takes
place.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2984
Fixes: 65122e9e8038 ("ir_constant_variable: New pass to mark constant-assigned variables constant.")
Reviewed-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5413>

radv: set DB_SHADER_CONTROL.CONSERVATIVE_Z_EXPORT correctly

Use the SPIR-V execution modes if set.

Cc: 20.1 <mesa-stable@lists.freedesktop.org>
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/5404>

android: nvir/gv100: update sources in Makefile.sources

Fixes the following building errors:

FAILED: out/target/product/x86_64/obj/SHARED_LIBRARIES/gallium_dri_intermediates/LINKED/gallium_dri.so
...
ld.lld: error: undefined symbol: nv50_ir::getTargetGV100(unsigned int)
...
ld.lld: error: undefined symbol: nv50_ir::getTargetGV100(unsigned int)
clang-9: error: linker command failed with exit code 1 (use -v to see invocation)

Fixes: 78103abe ("nvir/gv100: initial support")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>

freedreno/ir3: limit pre-fetched tex dest

Teach RA to setup additional interference to prevent textures fetched
before the FS starts from ending up in a register that is too high to
encode.

Fixes mis-rendering in multiple playcanv.as webgl apps.

Note that the regression was not actually 733bee57eb8's fault, but
that was the commit that exposed the problem.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3108
Fixes: 733bee57eb8 ("glsl: lower samplers with highp coordinates correctly")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5431>

freedreno/ir3: remove RA "q-values" optimization

This is mainly the "piglit optimization" (ie, since piglit launches an
separate process for for each test).  It was never wired up for a6xx,
and makes register class setup unnecessarily complicated.  Remove it to
simplify the next patch.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5431>

freedreno/ir3: respect tex prefetch limits

Refactor a bit the limit checking in the bindless case, and add tex/samp
limit checking for the non-bindless case, to ensure we do not try to
prefetch textures which cannot be encoded in the # of bits available.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5431>

freedreno/ir3: add debug code to print conflicting half-regs

I keep re-typing this from time to time when debugging various things.
Which is dumb.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5431>

nir/print: print tex dest type

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5431>

iris/icl+: Report same caching domain as main surface for clear color BO.

Even though the clear color BO is bound as a read-only buffer, report
the same caching domain as the main BO in use_surface() (typically
IRIS_DOMAIN_RENDER_WRITE) in order to avoid ping-ponging back and
forth between IRIS_DOMAIN_RENDER_WRITE and IRIS_DOMAIN_OTHER_READ,
which leads to increased stall-at-pixel-scoreboard synchronization
between draw calls.

Fixes a 5%-10% FPS regression in some benchmarks spotted on ICL.

Reported-by: Clayton Craft <clayton.a.craft@intel.com>
Fixes: eb5d1c27227302167d299 "iris: Annotate all BO uses with domain and sequence number information."
Closes: #3097
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5411>

android: aco: add aco_ir.cpp to Makefile.sources

Fixes the following building errors:

FAILED: out/target/product/x86_64/obj/SHARED_LIBRARIES/vulkan.radv_intermediates/LINKED/vulkan.radv.so
...
ld.lld: error: undefined symbol: aco::can_use_SDWA(chip_class, std::__1::unique_ptr<aco::Instruction, aco::instr_deleter_functor> const&)
...
ld.lld: error: undefined symbol: aco::can_use_opsel(chip_class, aco_opcode, int, bool)
...
clang-9: error: linker command failed with exit code 1 (use -v to see invocation)

Fixes: d9cfb8ad ("aco: validate instructions reading/writing upper halves/bytes")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5425>

docs: update calendar, add news item, and link releases notes for 20.1.1

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5421>

docs: Add release notes for 20.1.1

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5421>

ac/surface: don't free dcc_retile_map on failure

because the hash table now owns it.

Fixes: bd553f0546d - ac/surface: cache DCC retile maps (v2)
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5424>

ac/surface: enable DCC for the first level in the mip tail on gfx10

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5424>

ac/surface: require that gfx8 doesn't have DCC in order to be displayable

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5424>

ac/surface: don't set is_displayable if displayable DCC is missing

If flags.display isn't set, then displayable DCC will not be computed, so
is_displayable will always be false.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5424>

amd/addrlib: fix the C++ one definition rule violation

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/1854
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5414>

iris: Better handle metadata in NIR passes

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5171>

intel/nir: Call nir_metadata_preserve on !progress

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5171>

nir: Properly preserve metadata in more cases

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5171>

nir: Call nir_metadata_preserve on !progress

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5171>

nir: Add a nir_shader_preserve_all_metadata helper

There are some passes which really work on the shader level and it's
easier if we have a helper which preserves metadata on the whole shader.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5171>

nir: Add a nir_metadata_all enum value

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5171>

gallivm/sample: fix texel type for stencil 8-bit

This has to be unsigned, so clamping works properly for border
colors.

Fixes dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_uint_stencil

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5379>

gallivm/conv: enable conversion min code. (v2)

I'm not sure why this code was if (0), but if (1) for it fixes
dEQP-GLES31.functional.texture.border_clamp.range_clamp.nearest_float_color
This test expects +inf to get mapped to 255 and -inf to 0, both values
were ending up at 0.

v2: also enable in the SSE paths

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5379>

gallivm/format: convert unsigned values to float properly.

This fixes:
dEQP-GLES31.functional.draw_indirect.random.2

which ends up with 3x32-bit USCALED values going down this path
some of which have the top bit set, and end up converted to signed
float instead of unsigned float values.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5379>

llvmpipe: fix subpixel bits reporting.

This fixes some vulkan tests later.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5379>

gallivm/nir: add group barrier support

Fixes crash in
dEQP-GLES31.functional.synchronization.inter_invocation.image_write_read

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5379>

draw/gs: add more info to debugging.

adds invocations and vertex streams to default off debug,
fixes compile as well due to missing ,

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5379>

draw/gs: fix emitting inactive primitives crash

Fixes dEQP-GLES31.functional.geometry_shading.emit.line_strip_emit_1_end_1

This test only emits 1 primitive, but the stores don't respect
the current mask, which might only have one lane active, for that single
primitive. Also fix the final emit path to use the emitted_mask
rather than the current execution mask.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5379>

ci: Leave a note as to what might be going on with a test.

dEQP-GLES2.functional.clipping.triangle_vertex.clip_three.clip_neg_x_neg_z_and_pos_x_pos_z_and_neg_x_neg_y_pos_z
fails pretty strangely (given that we're passing everything else) and
there's an old VK-GL-CTS bug open about this test, and it's suspicious
that all the ARM drivers seem to have trouble with it.  I tried dropping
to -O0 on guilding that file in the CTS and it didn't help, though.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5419>

freedreno/a6xx: Fix clip_halfz support.

Same bit as on other gens, apparently it just got missed on this one.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5419>

nvc0: initial support for tu1xx

v2:
- add proper method definitions

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvc0: initial support for gv100

v2:
- remove unnecessary MAX2()
- add proper method definitions

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvc0: remove hardcoded blitter vertprog

I don't really feel like writing SM70 SASS by hand...

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvc0: move setting of entrypoint for a shader stage to a function

GV100 requires something different, cleaner to move this to a single place.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvc0: use NVIDIA headers for GP100- compute QMD

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvc0: use NVIDIA headers for GK104->GM2xx compute QMD

v2:
- add header debug_printf(), and indent the output
v3:
- rename one of the helper macros

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/gv100: enable support for tu1xx

SM75 has a bunch more stuff, but is otherwise backwards-compatible
with SM70 SASS.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/gv100: initial support

v2:
- add TargetGV100::isBarrierRequired() for OP_BREV
- use NV50_IR_SUBOP_LOP3_LUT() convenience macro where it makes sense
- separated out nir_lower_idiv into its own commit
- make use of the shared function to generate compiler options
- disable lower_fpow, nir's lowering is broken
v3:
- use replaceCvt() instead of custom NEG/ABS/SAT lowering
v4:
- remove WAR from peephole, not needed now we're using replaceCvt()

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir/gm107: switch off lower_extract_word

We can use PRMT here.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir/gm107: switch off lower_extract_byte

We can use PRMT here.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir/gm107: turn on nir_lower_extract64

About to disable lowering for extract_byte/word in favour of a better
local implementation, but still need lowering for 64-bit versions.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir/gm107: split nir shader compiler options from gf100

We can enable some more things here vs earlier GPUs.

v2:
- make use of the shared function to generate compiler options

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/gm107: separate out header for sched data calculator

SM70 code emitter will want to reuse this.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/gm107: replace SHR+AND+AND with PRMT+PRMT in PFETCH lowering

This is more SM70-friendly.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/gm107: implement OP_PERMT

PFETCH lowering will be changed to use this as it's more SM70-friendly,
and this will also allow us to implement extract_byte/word opcodes.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: use nir_lower_idiv

NIR provides a common implementation of this so we don't need to use a
hand-written built-in library.

v2:
- use idiv_precise instead

Especially important on SM70 where we don't have an assembler.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: nir expects the shift amount to wrap, rather than clamp

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: implement nir_op_uror

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: implement nir_op_urol

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: implement nir_op_extract_i16

v2:
- use getSSA() instead of getScratch()
v3:
- fix whitespace

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: implement nir_op_extract_u16

v2:
- use getSSA() instead of getScratch()
v3:
- fix whitespace

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: implement nir_op_extract_i8

v2:
- use getSSA() instead of getScratch()
v3:
- fix whitespace

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: implement nir_op_extract_u8

v2:
- use getSSA() instead of getScratch()
v3:
- fix whitespace

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: turn on lower_rotate

This isn't implemented, and won't be for GPUs that don't support SHF.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: flesh out options

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: move nir options to codegen

These seem to make more sense living with the compiler.

v2:
- use a shared function to generate the per-chipset structs
- remove nir.h include from header, not needed

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: fix fragment program output when using MRT

v2:
- use BITFIELD64_BIT()

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir/nir: use component helpers instead of insn->num_components

We have nir_intrinsic_dest_components and nir_intrinsic_src_components
which handle all the corner cases.

Fixes a bunch of regressions like front_face stuff.

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Ben Skeggs <bskeggs@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir: run replaceZero() before replaceCvt()

replaceCvt() will miss some cases otherwise.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir: add constant folding for OP_PERMT

Important for SM70 INSBF/EXTBF lowering, as these can can often be
eliminated completely.

v2:
- skip CF when subOp is set

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir: introduce OP_FINAL

Required to support SM70 GS.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir: introduce OP_SGXT

Required for SM70 EXTBF lowering.

v2:
- added constant folding

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>

nvir: introduce OP_BMSK

This replaces the existing implementation without adding lowering for
earlier GPUs.  The reason for this is because the existing code isn't
at all correct, and it also can't be hit anyway.

Will be required to support SM70 lowering passes.

v2:
- fixup source selection

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5377>