radv: don't lower indirects until after opts have run

Noticed while passing by. Not sure if it impacts anything, but
likely to impact GFX9 more than anything else since we lower
inputs, outputs and locals there.

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

st/nir: fix atomic lowering for gallium drivers

i965 and gallium handle the atomic buffer index differently. It was
just by luck that the single piglit test for this was passing.

For gallium we use the atomic binding so that we match the handling
in st_bind_atomics().

On radeonsi this fixes the CTS test:
KHR-GL43.shader_storage_buffer_object.advanced-write-fragment

It also fixes tressfx hair rendering in Tomb Raider.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/radeonsi: enable uniform packing in NIR backend

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st: add uniform packing support to lower_uniforms_to_ubo()

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

gallium: add packed uniform CAP

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/nir/radeonsi: move nir_lower_uniforms_to_ubo() to the state tracker

This will only ever be used by gallium drivers so it probably doesn't
belong in the nir toolkit. Also we want to pass it some non NIR
things in the following patch.

To avoid regressions we wrap the lowering calls that have been moved
to st_glsl_to_nir with a quick hack so that they are only called for
radeonsi, we will replace the hack with a check for uniform packing
in a following patch.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st: add st_glsl_type_dword_size() helper

This will be used to support uniform packing.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/glsl_to_nir: add support for packed builtin uniforms

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: add _mesa_add_sized_state_reference() helper

This will be used for adding packed builtin uniforms.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: add support propagate uniform support for packed uniforms

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: allow for uniform packing when adding uniforms to param list

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

mesa: add packing support for setting uniform handles

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

mesa: add packing support for setting uniforms

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

mesa: create copy uniform to storage helpers

These will be used in the following patch to allow copying directly
to the param list when packing is enabled.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: rework ParameterList to allow packing

Currently everything is padded to 4 components. Making the list
more flexible will allow us to do uniform packing.

V2 (suggestions from Nicolai):
- always pass existing calls to _mesa_add_parameter() true for padd_and_align
- fix bindless param value offsets
- remove left over wip logic from pad and align code
- zero out param value padding
- whitespace fix

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: add PackedDriverUniformStorage const

Will be used to determine whether to take packing code paths or not.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

broadcom/vc5: Don't annotate dumps with stale live intervals.

As you're debugging register allocation, you may have changed the
intervals and not recomputed yet.  Just skip the dump in that case.

broadcom/vc5: Add support for register spilling.

Our register spilling support is nice to have since vc4 couldn't at all,
but we're still very restricted due to needing to not spill during a TMU
operation, or during the last segment of the program (which would be nice
to spill a value of, when there's a long-lived value being passed through
with little modification from the start to the end).

We could do better by emitting unspills for the last-segment values just
before the last thrsw, since the last segment is probably not the maximum
interference area.

Fixes GTF uniform_buffer_object_arrays_of_all_valid_basic_types and 3
others.

broadcom/vc5: Remove redundant last_inst lookup.

The point was to get the MOV, which the MOV_dest already returned.

broadcom/vc5: On QPU pack error, dump the instruction and return cleanly.

This is nice for debugging when you've made a bad instruction.

broadcom/vc5: Add cursors to the compiler infrastructure, like NIR's.

This will let me do lowering late in compilation using the same
instruction builder as we use in nir_to_vir.

broadcom/vc5: Move the umul macro to a header.

Anywhere we want to multiply, we probably want this.

broadcom/vc5: Correct the arg count of TIDX/EIDX.

broadcom/vc5: Re-do live variables after removing thrsws.

Otherwise our start/ends ips won't line up with the actual instructions.

broadcom/vc5: Add a QPU helper for instructions using the TLB.

This will be used for detecting last thread segment in register spilling.

broadcom/vc5: Introduce v3d_qpu_reads_vpm()/v3d_qpu_writes_vpm().

These helpers will be used in register spilling to determine where to add
a last thrsw if needed, and might help refactor QPU scheduling.

broadcom/vc5: The ldvpm signal also a case of using the VPM.

The QPU scheduling code calling this function already separately checked
this signal.

broadcom/vc5: Extract v3d_qpu_writes_tmu() helper.

This will be reused in register spilling.

radv: don't export NULL layer.

We have some cases where in subpass we want the layer but having
it be 0 and loaded in the frag shader without the vertex shader
exporting it is fine.

So don't export the layer if we don't have a value to put in it.

Fixes: d4c74aed7a8 (radv/multiview: mark layer_input if we have input attachments.)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

mesa: adjust incorrect comment in texture_buffer_range

nir: Don't compare b2f or b2i with zero

All of the shaders that had loops changed were in Tomb Raider.  The one
shader that lost SIMD16 is one of those.

Skylake
total instructions in shared programs: 14391653 -> 14390468 (<.01%)
instructions in affected programs: 111891 -> 110706 (-1.06%)
helped: 501
HURT: 0
helped stats (abs) min: 1 max: 155 x̄: 2.37 x̃: 1
helped stats (rel) min: 0.05% max: 21.54% x̄: 1.61% x̃: 1.01%
95% mean confidence interval for instructions value: -3.23 -1.50
95% mean confidence interval for instructions %-change: -1.77% -1.45%
Instructions are helped.

total cycles in shared programs: 532793024 -> 532776598 (<.01%)
cycles in affected programs: 987682 -> 971256 (-1.66%)
helped: 348
nnHURT: 41
helped stats (abs) min: 1 max: 3074 x̄: 54.91 x̃: 18
helped stats (rel) min: 0.05% max: 32.24% x̄: 3.36% x̃: 1.68%
HURT stats (abs)   min: 1 max: 422 x̄: 65.39 x̃: 24
HURT stats (rel)   min: 0.09% max: 39.29% x̄: 9.50% x̃: 2.02%
95% mean confidence interval for cycles value: -64.08 -20.38
95% mean confidence interval for cycles %-change: -2.78% -1.23%
Cycles are helped.

total loops in shared programs: 4854 -> 4829 (-0.52%)
loops in affected programs: 27 -> 2 (-92.59%)
helped: 18
HURT: 0

LOST:   1
GAINED: 0

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

radv: lower constant initializers on output variables earlier

If a shader only writes to an output via a constant initializer we
need to lower it before we call nir_remove_dead_variables so that
this pass sees the stores from the initializer and doesn't kill the
output.

Fixes test failures in new work-in-progress CTS tests:
dEQP-VK.spirv_assembly.instruction.graphics.variable_init.output.float

This is ported from anv:
99b57daf4a anv/pipeline: lower constant initializers on output variables earlier
from Iago Toral Quiroga <itoral@igalia.com>

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

radv/query: handle multiview timestamp queries.

For each view bit we need to emit a timestamp query.

Fixes: dEQP-VK.multiview.queries*
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

radv/query: handle multiview queries properly. (v3)

For multiview we need to emit a number of sequential queries
depending on the view mask.

This avoids dEQP-VK.multiview.queries.15 waiting forever
on the CPU for query results that are never coming.

We only really want to emit one query,
and the rest should be blank (amdvlk does the same),
so we emit begin/end pairs for all the others except
the first query.

v2: fix tests
v3: split out patch.

Fixes: dEQP-VK.multiview.queries*
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

radv/query: split out begin/end query emission

This just splits out the begin/end query hw emissions,
it makes it easier to add multiview support for queries.

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

radv/multiview: mark layer_input if we have input attachments.

This fixes:
dEQP-VK.multiview.input_attachments*

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

anv/pipeline: set active_stages early

Since the intermediate states of active_stages are not used,
i.e. active_stages is read only after all stages were set into it,
just set its value before compiling the shaders.

This will allow to conditionally run certain passes based on what
other shaders are being used, e.g. a certain pass might only be
applicable to the vertex shader if there's no geometry or tessellation
shader being used.

v2: Use vk_to_mesa_shader_stage. (Lionel)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

anv/pipeline: fail if TCS/TES compile fail

v2: Add Fixes tag. (Lionel)

Fixes: e50d4807a35e679 ("anv: Compile TCS/TES shaders.")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

main/program_binary: In ProgramBinary set link status as LINKING_SKIPPED

This change allows the disk shader cache to work with programs loaded
with ProgramBinary. Drivers check for LINKING_SKIPPED, and if set,
then they try to use the shader cache.

Since the program loaded by ProgramBinary is similar to loading the
shader from the disk cache, this is probably more appropriate.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

i965: Allow disk shader cache usage with LINKING_SUCCESS status

Currently, we only look in the disk shader cache if we see that the
shader program is in the cache during the link step.

If the shader cache entry isn't found during the program link, there
are still some (fairly unlikely) scenarios where later it might be
useful to search the cache for gen binary programs.

1. If the cache evicts the serialized glsl cache, there might still be
   valid gen program entries in the disk cache.

2. If two applications are running in parallel, then it is possible
   that one may write out the cached gen program item which the other
   application can then make use of.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

glsl/serialize: Save shader program metadata sha1

When the shader cache is used, this can be generated. In fact, the
shader cache uses this sha1 to lookup the serialized GL shader
program.

If a GL shader program is restored with ProgramBinary, the shaders are
not available, and therefore the correct sha1 cannot be generated. If
this is restored, then we can use the shader cache to restore the
binary programs to the program that was loaded with ProgramBinary.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

glsl: Remove api_enabled tracking for transform feedback

We used this to prevent usage of the disk shader cache when transform
feedback was enabled via the GL API. This is no longer used.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105444
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

i965: Allow disk shader cache usage with transform feedback

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105444
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

glsl/shader_cache: Allow shader cache usage with transform feedback

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105444
Suggested-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

scons: need to split CC or things might fail

We've seen this fail internally.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>

i965: Add INTEL_DEBUG stages support for disk shader cache

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

radv: handle exporting view index to fragment shader. (v1.1)

The fragment shader was trying to read this, but nothing
was exporting it from the vertex shader. This handles
it like the prim id export.

Fixes:
dEQP-VK.multiview.secondary_cmd_buffer.*
dEQP-VK.multiview.index.fragment_shader.*

v1.1: updated to use 0x1 (Samuel)

Fixes: e3265c10c89 (radv: Implement multiview draws.)
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

st/nine: Fix non inversible matrix check

There was a missing absolute value when
checking if the determinant was big enough.

Fixes: https://github.com/iXit/Mesa-3D/issues/292
Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
CC: "17.3 18.0" <mesa-stable@lists.freedesktop.org>

st/nine: Fixes warning about implicit conversion

Makes the conversion explicit.

Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=102542
Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
CC: "17.3 18.0" <mesa-stable@lists.freedesktop.org>

st/nine: Fix bad tracking of vs textures for NINESBT_ALL

Stateblocks with NINESBT_ALL should track all textures.
For better performance they have a faster path which
copies all the required.

This path was only tracking ps textures.

Fixes: https://github.com/iXit/Mesa-3D/issues/303
Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
CC: "17.3 18.0" <mesa-stable@lists.freedesktop.org>

st/nine: Fix bad tracking of bound vs textures

An incorrect formula was used to compute bound_samplers_mask_vs.
Since s is above always 8 for vs and the variable is encoded on 8 bits,
it was always 0.
This resulted in commiting the samplers every call when
there was at least one texture read in the vs shader.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

radv: make vk_format_description structures static

No need to bother the linker about them.

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

radv: fix stale comment in generated vk_format_table.c

It seems to be a leftover from u_format_table.py.

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

anv: Silence warning about heap_size.

We only get VK_SUCCESS if it was initialized, but apparently my compiler
doesn't track that far.

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

i965: Silence compiler warning about promoted_constants.

We only have a cfg != NULL if we went through one of the paths that set
it, but my compiler doesn't figure that out.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 6411defdcd6f ("intel/cs: Re-run final NIR optimizations for each SIMD size")

anv: Silence compiler warnings about uninitialized bind_offset.

This is a legitimate warning: if anv's blorp_alloc_binding_table() throws
an error from anv_cmd_buffer_alloc_blorp_binding_table(), we silently
continue to use this undefined value.  The rest of this code doesn't seem
very allocation-error-proof, though, either.

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

intel/compiler: Use gen_get_device_info() in test_eu_validate

Previously the unit test filled out a minimal devinfo struct. A previous
patch caused the test to begin assert failing because the devinfo was
not complete. Avoid this by using the real mechanism to create devinfo.

Note that we have to drop icl from the table, since we now rely on the
name -> PCI ID translation done by gen_device_name_to_pci_device_id(),
and ICL's PCI IDs are not upstream yet.

Fixes: f89e735719a6 ("intel/compiler: Check for unsupported register sizes.")
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

intel: Add cfl to gen_device_name_to_pci_device_id()

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

meson+dri3: allow building against older xcb (v3)

Similar to previous patch, make xcb 1.13 optional.

Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

dri3: allow building against older xcb (v3)

I'm not sure everyone wants to be updating their dri3 in a forced
march setting, this allows a nicer approach, esp when you want
to build on distro that aren't brand new.

I'm sure there are plenty of ways this patch could be cleaner,
and I've also not built it against an updated dri3.

For meson I've just left it alone, since if you are using meson
you probably don't mind xcb updates, and if you are using meson
you can fix this better than me.

v3: just don't put a version in for dri3/present without
modifiers, should allow building with 1.11 as well

(feel free to supply meson followups)

Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

r600: consolidate PIPE_BIND_SHARED/SCANOUT handling

(Ported from radeonsi commit f70f6baaa3bb0f8b280ac2eaea69bbffaf7de840)

Allows cached BOs to be reused in more cases.

Bugzilla: https://bugs.freedesktop.org/105171
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>

intel/compiler: Check for unsupported register sizes.

Make sure we don't emit 64 bit types if the hardware doesn't support
them.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

loader: Include include/drm-uapi in the autotools build

We're already including it in the meson build.  This fixes build issues
on systems which have a drm_fourcc.h that doesn't have modifiers.

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

egl/android: Implement the eglSwapinterval for Android.

Implement the eglSwapinterval for Android platform to
enable the async mode for some GFX benchmarks such as
Daimler C217, CityBench.

Results of the dEQP-EGL.*swap_interval tests

'dEQP-EGL.functional.query_config.get_config_attrib.max_swap_interval'..
'dEQP-EGL.functional.query_config.get_config_attrib.min_swap_interval'..
'dEQP-EGL.functional.choose_config.simple.selection_only.max_swap_interval'..
'dEQP-EGL.functional.choose_config.simple.selection_only.min_swap_interval'..
'dEQP-EGL.functional.choose_config.simple.selection_and_sort.max_swap_interval'..
'dEQP-EGL.functional.choose_config.simple.selection_and_sort.min_swap_interval'..
'dEQP-EGL.functional.negative_api.swap_interval'..

 Test run totals:
   Passed:        7/7 (100.0%)
   Failed:        0/7 (0.0%)
   Not supported: 0/7 (0.0%)
   Warnings:      0/7 (0.0%)

Signed-off-by: Zhongmin Wu <zhongmin.wu@intel.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
[Emil Velikov: polish inline comment, add dEQP stats, s/dpy/disp/]
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

st/mesa: simplify st_init_limits() via tgsi_processor_to_shader_stage

Reuse the tgis helper and remove a bunch of duplicated code.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

tgsi: move tgsi_processor_to_shader_stage() to a header

This way we can utilise it with later patches.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

egl/dri2: move wayland header inclusion where applicable

Instead of indirectly pulling the wayland headers everywhere, use
forward declarations and #include only as needed.

Should effectively fix build errors like the following:

make[5]: Entering directory
'/.../src/gallium/state_trackers/omx/tizonia'
   CC       h264dprc.lo
In file included from h264dprc.c:45:0:
.../src/egl/drivers/dri2/egl_dri2.h:47:10: fatal error:
wayland/wayland-egl/wayland-egl-backend.h: No such file or directory
  #include "wayland/wayland-egl/wayland-egl-backend.h"

Cc: Dylan Baker <dylan@pnwbakers.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Eric Engestrom <eric.engestrom@imgtec.com>
Tested-by: Andy Furniss <adf.lists@gmail.com>

vulkan/wsi/x11: correct DRI3 version in comment

During development the version was bumped, yet the comment did not get
an update.

Fixes: c80c08e2260 ("vulkan/wsi/x11: Add support for DRI3 v1.2")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>

vulkan/wsi/x11: use ARRAY_SIZE where applicable

Use the handy macro instead of hard coded numbers.

Fixes: c80c08e2260 ("vulkan/wsi/x11: Add support for DRI3 v1.2")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>

mesa: RGB9_E5 invalid for CopyTexSubImage* in GLES

According to OpenGL ES 3.2, section 8.6, CopyTexSubImage* should return
an INVALID_OPERATION if the internalformat of the texture is RGB9_E5.

This fixes
dEQP-GLES31.functional.debug.negative_coverage.*.copytexsubimage2d_texture_internalformat.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

etnaviv: remove superfluous \n from DBG(..) callers

The DBG(..) macro appends a \n already so there is no
need to do it twice.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>

radv: run nir_opt_move_load_ubo

Polaris10:
SGPRS: 108560 -> 107856 (-0.65 %)
VGPRS: 74576 -> 74520 (-0.08 %)
Spilled SGPRs: 7375 -> 7113 (-3.55 %)
Code Size: 4273464 -> 4274364 (0.02 %) bytes
Max Waves: 9434 -> 9446 (0.13 %)

Vega10:
Totals from affected shaders:
SGPRS: 108264 -> 107576 (-0.64 %)
VGPRS: 69068 -> 69000 (-0.10 %)
Spilled SGPRs: 7221 -> 6959 (-3.63 %)
Code Size: 3800796 -> 3801496 (0.02 %) bytes
Max Waves: 10687 -> 10709 (0.21 %)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

nir: add nir_opt_move_load_ubo() optimization pass

This pass moves load UBO operations just before their first use,
loosely based on nir_opt_move_comparisons.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

radv: drop geometry stride user sgpr.

This removes the other geometry specific user sgpr.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: get rid of geometry user sgpr for num entries.

This drops one of the geometry specific user sgprs,
we can work this out at compile time.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: migrate lds size calculations to shader gen.

This moves the lds_size calcs into the shader so we have all
the size stuff in one file.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: drop scanning the tess shader in the nir code.

This drops the now unneeded scanning and results in favour
of the ones in the info.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: use num_patches output from tcs shader.

Instead of recalculating the value, use the shader calculated value.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/tess: remove last chunk of tess sgprs

This removes the last TES-specifc user sgpr.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: pass num_patches to tes from tcs

TES needs num_patches to do some of the calculations.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: drop tess offchip layout for tcs.

This removes the last TCS specific user sgpr.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: drop tcs_out_offsets

Move all calculations to shader generation.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: drop tcs_out_layout

Move all calculations to shader generation.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/tess: drop tcs_in_layout setting completely.

Inline all calcs at shader creation.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: drop ls_out_layout const.

We can precalculate input_vertex_size at compile time.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/shader_info: start gathering tess output info (v2)

This gathers the ls outputs written by the vertex shader,
and the tcs outputs, these are needed to calculate certain
tcs parameters.

These have to be separate for combined gfx9 shaders.

This is a bit pessimistic compared to the nir pass,
as we don't work out the individual slots for tcs outputs,
but I actually thing it should be fine to just mark the whole
thing used here.

v2: move to radv, handle clip dist (Samuel),
    handle compacts and patchs properly.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: migrate unique index info shader info (v2)

This just moves this function to an inline so the shader_info
pass can use it.

v2: use inline (Samuel)

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

Revert "mesa: do not trigger _NEW_TEXTURE_STATE in glActiveTexture()"

This reverts commit f314a532fdc7af8381586144d2631d9968331f05.

This appears to introduce some blinking textures in UT2004. Not
sure exactly what's the root cause because we don't have much
information about the issue.

Anyway, this was just a micro optimization that actually breaks,
at least, one app almost one year later.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105436
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

anv: silence unused variable warning

Fixes: 59b0ea0c748 ("anv: Stop returning VK_ERROR_INCOMPATIBLE_DRIVER")
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

i965: silence unused function warning

[123/227] Compiling C object 'src/mesa/drivers/dri/i965/libi965_gen110@sta/genX_blorp_exec.c.o'.
../src/mesa/drivers/dri/i965/genX_blorp_exec.c:99:1: warning: ‘blorp_get_surface_base_address’ defined but not used [-Wunused-function]
 blorp_get_surface_base_address(struct blorp_batch *batch)
 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

anv: silence unused function warning on gen11

[84/227] Compiling C object 'src/intel/vulkan/libanv_gen110@sta/genX_blorp_exec.c.o'.
../src/intel/vulkan/genX_blorp_exec.c:68:1: warning: ‘blorp_get_surface_base_address’ defined but not used [-Wunused-function]
 blorp_get_surface_base_address(struct blorp_batch *batch)
 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

meson: fix pipe-loaders after omx changes

with_gallium_omx used to be a boolean, but now it's a string. That means
it needs to be compared to 'disabled' instead of false.

CC: Rob Clark <robdclark@gmail.com>
Fixes: 34e852d5b50772199797ea839fc8d6b3805633ff
       ("meson: Re-add auto option for omx")
Signed-off-by: Dylan Baker <dylan.c.baker@intel.com>
Tested-by: Rob Clark <robdclark@gmail.com
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

meson: require amdgpu >= 2.4.91

the meson equivalent of f8773edb0a0cf5ff885cd3494717323983eedeed

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

configure.ac: require libdrm_amdgpu 2.4.91

Since 2.4.90 is problematic, just ask for the next version.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

configure.ac: blacklist libdrm 2.4.90

Cc: 18.0 17.3 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

radv: dump LLVM IR when a hang is detected

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

radv: record LLVM IR when debugging shaders

If AMD_shader_info or RADV_TRACE_FILE is used we might need to
keep trace of LLVM IR.

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

radv: add dump_shader to the NIR compiler options

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

radv: pass the NIR compiler options to ac_compile_llvm_module()

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

radv: print some information when RADV_TRACE_FILE is set

Just to be sure all options are enabled when trying to generate
a hang report.

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