intel/tools/aub_dump: move aub file initialization to maybe_init()

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3705>

lima: expose fragment shader derivatives capability

Support for fragment shader derivatives has landed in the Lima PP
compiler for a long time, but its capability is not exposed yet.

Expose the support now.

Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3944>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3944>

v3d: Sync on last CS when non-compute stage uses resource written by CS

When a resource is written by a compute shader and then used by a
non-compute stage we sync on last compute job to guarantee that the
resource has been completely written when the next stage reads resources.

In the other cases how flushes are done guarantee the serialization of
the writes and reads.

To reproduce the failure the following tests should be executed in batch
as last test don't fail when run isolated:

KHR-GLES31.core.shader_image_load_store.basic-allFormats-load-fs
KHR-GLES31.core.shader_image_load_store.basic-allFormats-loadStoreComputeStage
KHR-GLES31.core.shader_image_load_store.basic-allTargets-load-cs
KHR-GLES31.core.shader_image_load_store.advanced-sync-vertexArray

v2: Use fence dep instead of bo_wait (Eric Anholt)
v3: Rename struct names (Iago Toral)
    Document why is not needed on graphics->compute case. (Iago Toral)
    Follow same code pattern of the other update of in_sync_bcl.
v4: Fixed comments style. (Iago Toral)

Fixes KHR-GLES31.core.shader_image_load_store.advanced-sync-vertexArray

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
CC: 19.3 20.0 <mesa-stable@lists.freedesktop.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2700>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2700>

gitlab-ci: Enable the lima job again

Flaky tests should be fixed to the best of our knowledge.
Fails and skips lists should be up-to-date again.

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3884>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3884>

gitlab-ci: lima: Add flaky tests to the skips list

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Cc: <mesa-stable@lists.freedesktop.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3884>

nir: fix gl_nir_lower_images for bindless images

Fixes: 7342b859afb5a7e7f9fb1813e7ab3a55a1c8a704
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3938>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3938>

freedreno/computerator: fix build dependency

Ensure the generated register headers are built before computerator uses
them.

Reported-by: Clayton Craft <clayton.a.craft@intel.com>
Signed-off-by: Rob Clark <robdclark@chromium.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3939>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3939>

glx/drisw: fix shm put image fallback

The fallback to the non-shm put path used the wrong width here
as the pixmap is still allocated in a shared segment, so the
width needs to reflect that.

Fixes: 02c3dad0f3b4 ("Call shmget() with permission 0600 instead of 0777")
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3823>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3823>

glx/drisw: return false if shmid == -1

If an attempt to create an shm pixmap in XCreateDrawable fails
then it ends up with the shmid == -1. This means the get image
path needs to fallback so return false in this case to use the
non-shm get image path.

Fixes: 02c3dad0f3b4 ("Call shmget() with permission 0600 instead of 0777")
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3823>

glx/drisw: add getImageShm2 path

This adds return values to the get image path, so the caller can fallback.

Fixes: 02c3dad0f3b4 ("Call shmget() with permission 0600 instead of 0777")
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3823>

dri: add another get shm variant.

When Brian in 02c3dad0f3b4d26e0faa5cc51d06bc50d693dcdc restricted
the shm permissions it means we hit the fallback paths in some
scenarios we hadn't before.

When you use Xephyr to xdmcp from one user to another the new perms
stop the X server (running as user a) attaching to the SHM segments
from gnome-shell (running as user b).

In this case however only the GLX side of the code had insight into this,
and the dri could was meant of fall back, and it worked for put image
fine but the get image path was broken, since there was no indication
in the broken case of the need to fallback.

This adds a return type to a new interface member that lets the
caller know it has to fallback.

Fixes: 02c3dad0f3b4 ("Call shmget() with permission 0600 instead of 0777")
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3823>

ci: Blacklist another freedreno flaky test.

This is the recurring flake from the last week, including spuriously
failing a pipeline once.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3937>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3937>

intel/isl: Add isl_aux_info.c to Makefile.sources

This should fix the Android build.

Fixes: 58d4749e56 "isl: Add a module which manages aux resolves"
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reported-by: Clayton Craft <clayton.a.craft@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3934>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3934>

intel/blorp: Implement GEN:BUG:1605967699.

v2:
 - Update comments and refactor code (Lionel).
 - Only apply workaround to stencil resolves.

Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3909>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3909>

gallium/util: remove unused debug_print_foo helpers

These are unused, so let's just get rid of them.

Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3901>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3901>

gallium/util: do not use debug_print_format

These are the only two places we use this macro, and it's no longer very
gallium-specific. So let's get rid of this, and just use debug_printf
and util_format_name directly instead.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3901>

util: move debug_memory_{begin,end} to os_memory_debug.h

This is where the other debug_memory_* functions are declared, so let's
move it here for symmetry.

This allows us to drop an include of u_debug_gallium.h, which makes us
depend on gallium-headers in non-gallium code.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3901>

hud: add GALLIUM_HUD_SCALE

Scale hud by an integer factor, for high DPI displays.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3931>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3931>

turnip: move tile_load_ib/sysmem_clear_ib into draw_cs

Avoids having to calculate reserved sizes for substream cs, also matches
what the blob does.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3925>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3925>

turnip: make cond_exec helper easier to use

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3925>

turnip: remove marker seqno

Use robclark's new crashdec/devcoredump thing instead.

Note: not sure this ever really worked because it didn't WFI.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3925>

turnip: automatically reserve cmdstream space in emit_pkt4/emit_pkt7

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3925>

turnip: add tu_device pointer to tu_cs

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3925>

turnip: fix COND_EXEC reserved size in tu_query

Conditionally executed dwords must be in the same bo.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3925>

freedreno/computerator: add computerator

A standalone tool to compile and run compute shaders from ir3 assembly.
Mostly to have an easy way to experiment with instructions.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3926>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3926>

freedreno/ir3: allow block->predecessors to be null

This way we can also use ir3_print from computerator, which mostly
bypasses the ir3_block construct (since it doesn't need to do
scheduling, etc)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3926>

freedreno/computerator: rename prefix asm->ir3

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3926>

freedreno/computerator: polish out some of the rust

Updates for differences between fdre-a3xx's early version of ir3, and
what we have now in mesa.  And updates for instruction name and syntax
changes.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3926>

freedreno/computerator: import parser/lexer from fdre-a3xx

Import the rusty old parser from freedreno.git

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3926>

lima: remove its hash table entry when invalidating a resource

When a resouce is already invalidated, its hash table entry becomes
useless. In addition, the lima_job_free() function won't remove the hash
table entry for invalidated resource. So the hash entry should be
removed when invalidating the resource, otherwise bogus hash entry might
be left in the table, and when the resource is reused in another job,
the code will find the freed job when invalidating and thus result in crash.

Fixes: c64994433c0d ("lima: track write submits of context (v3)")
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3917>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3917>

nir, intel: Move use_scoped_memory_barrier to nir_options

This option will be used later by GLSL, so move to a common struct.

Because nir_options is filled in the compiler instead of the Vulkan
driver, fix that up.  GLSL will ignore that for now.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3913>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3913>

nir/tests: Use nir_scoped_memory_barrier() helper

Most of the vars tests already had a local helper, so just drop it in
favor of the one in nir_builder.  Remaining two tests changed to use
the helper.

The load_store_vectorizer tests were using the specific memory
barriers, but since scoped barriers are also handled, prefer that.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3913>

nir: Add the alias NIR_MEMORY_ACQ_REL

This will help upcoming C++ code that will have to combine those two
semantics.  In C++ it is not possible to do this without a cast or
adding an operator| to the enum.  Since having the short form will
also be convient to C, we picked the former solution.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3913>

nir/builder: Add nir_scoped_memory_barrier()

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3913>

freedreno: Switch to using lowered image intrinsics.

This cuts out a bunch of deref chain walking that the compiler can do for
us.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

nir: Move intel's intrinsic_image_coordinate_components() to core nir.

This is a query that both Intel and freedreno need to do.  We can simplify
it a lot with the new glsl_get_sampler_dim_coordinate_components()

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

freedreno/ir3: Fix the arg to ir3_get_num_components_for_image_format()

GLuint worked fine for storing our enum, but it should be an enum
pipe_format since the image-formats merge.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

prog_to_nir: Reuse glsl_get_sampler_dim_coordinate_components().

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

tgsi_to_nir: Reuse glsl_get_sampler_dim_coordinate_components().

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

freedreno/ir3: Reuse glsl_get_sampler_dim_coordinate_components() in tex_info.

Now that we have access to the interior switch statement not going through
the txs special case for coord_components, we can just use it.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

spirv_to_nir: Reuse glsl_sampler_dim_coordinate_components().

We just needed to move the SUBPASS_MS case in, and the rest of the cases
match up.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

glsl: Factor out the sampler dim coordinate components switch statement.

I want to reuse this in NIR image intrinsics in backends, which just have
dim/is_array.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

v3d: Ask the state tracker to lower image accesses off of derefs.

This saves a bunch of hassle in handling derefs in the backend, and would
be needed for reasonable handling of dynamic indexing of image arrays.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

gallium: Add a cap for enabling lowering of image load/store intrinsics.

The deref stuff is hard to handle in a backend supporting dynamic
indexing, while the lowering can easily turn that into the same kind of
dynamic indexing we do for textures, UBOs, and SSBOs.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

nir: Make image lowering optionally handle the !bindless case as well.

iris was doing this internally, but let's rename the function and move the
iris code there.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

nir: Rename gl_nir_lower_bindless_images.c in preparation for extending it.

The bulk of it can be reused to implement iris's internal non-bindless
image lowering, which I would like to reuse in freedreno, v3d, and
nir-to-tgsi.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3728>

i965: Use isl_aux_state_transition_write()

v2. Dirty shadow miptrees independent of aux. (Jason)

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2957>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2957>

i965: Use ISL's access preparation functions

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2957>

iris: Use isl_aux_state_transition_write()

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2957>

iris: Use ISL's access preparation functions

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2957>

iris: Use isl_aux_usage_has_fast_clear()

Make sure fast-clears aren't attempted or allowed for ISL_AUX_USAGE_MC.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2957>

isl: Add a module which manages aux resolves

Provide a generic interface which manages aux resolves in ISL. The
feature differences between this and what's in iris is:
* Support for media compression. ISL_AUX_USAGE_MC behaves differently
  from many other usages of CCS, so it was useful to implement this
  support upfront, while designing the interfaces.
* Optimizations for full-surface writes. For example, after a
  full-surface write occurs with ISL_AUX_USAGE_CCS_E in the PARTIAL_CLEAR
  state, isl_aux_state_transition_write() returns COMPRESSED_NO_CLEAR
  instead of COMPRESSED_CLEAR.

A performance suggestion for main-surface-invalidating/replacing writes
is given as a comment instead of adding a boolean to
isl_aux_prepare_access(). This avoids extra validation and should be
simple enough for the caller to handle.

v2. Add assertions. (Jason)
v3. Use switches in 2 more functions. (Jason)
    Store aux metadata in a static table. (Jason)
    Change prepare and finish function signatures. (Jason)
    Keep isl_aux_state_transition_* functions separate.
v4. (Jason)
    Assert against resolving in AUX_INVALID.
    Rename aux_info struct to aux_usage_info.
    Drop the justification for each aux_usage_info field.
    Split out the NONE case in write function.
    Restructure tests to more easily confirm coverage.
    Rename access_compressed field to compressed.
    Make write behavior less ambiguous.
v5. (Jason)
    Add more detail above WRITES_RESOLVE_AMBIGUATE.
    Add ISL_AUX_USAGE_MC to WritesResolveAmbiguate.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2957>

freedreno/ir3: Lower output precision

This lowers mediump FS outputs to fp16 in the ir3 backend. For now
this is a modest improvement, which mostly helps us whittle down the
full mediump work.  Once the GLSL level support lands, then right hand
side of the store output intrinsics will be fp16 expressions and we'll
cancel out the fp16 -> fp32 -> fp 16 round trip here.

We've had different attempts at implementing this: rewriting stores in
the GLSL IR, lowering GLSL IR outputs to temporaries and inserting
conversions when writing the temporaries to the outputs.  In the end,
GLSL ends up getting in the way a lot and doing it at the nir level is
easier and still possible since we have the output var precisions.

This part of the fp16 work is more of a step on the way towards full
fp16 support and will add a few extra conversion instructions:

total instructions in shared programs: 8151 -> 8163 (0.15%)
instructions in affected programs: 1187 -> 1199 (1.01%)
helped: 4
HURT: 10

total nops in shared programs: 3146 -> 3152 (0.19%)
nops in affected programs: 563 -> 569 (1.07%)
helped: 5
HURT: 10

total non-nops in shared programs: 5005 -> 5011 (0.12%)
non-nops in affected programs: 92 -> 98 (6.52%)
helped: 0
HURT: 3

total dwords in shared programs: 12832 -> 12800 (-0.25%)
dwords in affected programs: 96 -> 64 (-33.33%)
helped: 1
HURT: 0

total last-baryf in shared programs: 118 -> 115 (-2.54%)
last-baryf in affected programs: 21 -> 18 (-14.29%)
helped: 1
HURT: 0

total full in shared programs: 424 -> 417 (-1.65%)
full in affected programs: 15 -> 8 (-46.67%)
helped: 7

HURT: 0
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3822>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3822>

nir/types: Add glsl_float16_type() helper

This returns the float16 version of a float type.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3822>

freedreno/ir3: handle half registers for arrays during register allocation.

So far we only handle full regs of arrays during pre-allocation.
This patch is to handle half regs of arrays and also consider the size
of half regs when finding out conflicts.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3822>

nir: Add optimization for doing removing f16/f32 conversions

This eliminates conversions between f16 and f32 where possible. We can
always remove an upcast followed by a down cast, that is:

  f2f16 ( f2f32 (a) )  ->  a
  f2fmp ( f2f32 (a) )  ->  a

In the other direction, f2f16 loses precision and can't be undone by a
f2f32.  However, by definition it's always safe to elminate f2fmp:

  f2f32 ( f2fmp (a) )  ->  a

v2. [Neil Roberts (nroberts@igalia.com)]

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3822>

freedreno/ir3: Add new ir3 pass to fold out fp16 conversions

This pass tries to fold f2f16 conversion into alu instructions.
This will be useful to help reduce the number of instructions once
mesa starts supporting precision lowering.  For example:

  add.f r0.w, r0.w, c0.x
  cov.f32f16 hr2.x, r0.w

to

  add.f hr2.x, r0.w, c0.x

Additionally this pass also tries to fold f2f16 conversion into load_input
instruction:

  bary.f r0.x, 3, r0.w
  cov.f32f16 hr0.x, r0.x

to

  bary.f hr1.x, 3, r0.x

v2: Edit to not fold OPC_MAX_F and OPC_MIN_F, since that's not valid.

v3: Add OPC_ABSNEG_F to the blacklist as well.

v4: Don't remove dead cov instructions, DCE will do that later; don't
iterate through sources when a cov only has one; remove special
handling of IR3_REG_ARRAY and IR3_REG_RELATIV.

v5: Handle folding into u32.u32 movs of floats correctly, don't bail
out on IR3_REG_RELATIV or IR3_REG_ARRAY movs.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3822>

nir/opcodes: Add nir_op_f2fmp

This opcode is the same as the f2f16 opcode except that it comes with
a promise that it is safe to optimise it out if the result is
immediately converted back to a 32-bit float again. Normally this
would be a lossy conversion and so it would be visible to the
application, but if the conversion is generated as part of the mediump
lowering process then this removal doesn’t matter. The opcode is
eventually replaced with a regular f2f16 in the late optimisations so
the backends don’t need to handle it.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3822>

glapi/copyimage: Implement CopyImageSubDataNV

Implement CopyImageSubDataNV from NV_copy_image spec.
This is derived out of the existing implementation of CopyImageSubData.
It differs from CopyImageSubData in accordance with the differences
laid down in the ARB_copy_image spec.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3649>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3649>

iris: Fix import sync-file into syncobj

When importing a sync-file, the kernel expects to be told which syncobj
to replace with the new fence -- it does not automatically create a new
handle for us. Abide by this rule and create a new syncobj for the
imported sync-file.

Fixes: f459c56be6bf ("iris: Add fence support using drm_syncobj")
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3919>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3919>

pan/midgard: Implement load/store_shared

Shared memory is implemented almost identically to global memory from an
ISA perspective, so let's handle the new intrinsics. We include a code
path for constant offsets, which doesn't come up for globals.

Fixes dEQP-GLES31.functional.compute.basic.shared_var_single_invocation

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3775>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3775>

pan/midgard: Implement nir_intrinsic_get_buffer_size

We route it as a sysval. Fixes dEQP-GLES31.functional.compute.basic.ssbo_unsized_arr_single_invocation

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3775>

pan/midgard: Lower SSBOs in NIR

We need to lower SSBOs to globals regardless. Rather than do this in our
backend like we do now, use the common NIR pass, which will support
bounds checking.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3775>

turnip/pipeline: Don't assume tu_shader is a valid object

Fixes a crash in tu6_emit_fs_config() when 'shader' argument is
assumed to be non-null, which is possible.

Fixes dEQP test:

dEQP-VK.api.descriptor_set.descriptor_set_layout_lifetime.graphics

Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3756>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3756>

radv: add the trace BO to the BO list at submit time

Instead of adding it in every command buffer.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3891>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3891>

gallium/swr: Fix min/max range index draw

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3905>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3905>

iris: Set MOCS for constant packets on Gen12+

It seems to be back, and we shouldn't use 0, as that's now considered
an error.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3720>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3720>

iris: Fix BLORP vertex buffers to respect ISL MOCS settings

Fixes: a4da6008b6a ("iris: Use mocs from isl_dev.")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3720>

iris: Make mocs an inline helper in iris_resource.h

Now that it uses ISL rather than genxml code, there's no need for it to
live as a vtable function inside the state module.  We can just make it
a static inline helper in iris_resource.h so it's available throughout
the codebase.

Fixes: a4da6008b6a ("iris: Use mocs from isl_dev.")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3720>

ci: Remove a useless filtering of the lava logs.

We don't print every case any more, so no need to filter them out.  This
makes it so the output form "lavacli jobs logs" gets line-buffered into
"tee" and you can actually see what happened when the job is stuck but
before it times out.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3883>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3883>

ci: Don't bother generating deqp junit results since we don't present it.

We disabled presentation a while back because it's so expensive for gitlab
to parse it on the other side.  We may have a use for it some day if
gitlab gets better, but for now let's not spend the time processing it.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3883>

ci: Document how LAVA runners work.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3883>

ci: Make LAVA job fails emit the full list of unexpected test results.

When bringing up a new board or starting a new GLES version, we have a lot
of unexpected fails to document, so we need the full list in the log (not
just deqp-runner.sh's head -n 50) so we can populate the xfail list.

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3883>

ci: Make sure that we have a proper shell prompt for LAVA.

LAVA finds a '#' early in boot and races to emit its shell commands.
Apparently for the current boards those serial commands end up getting
buffered such that things work out, but for db410c and db820c, the buffer
is lost and LAVA gets stuck waiting for the prompt.  By setting a prompt,
we can delay our commands until we're actually supposed to emit them (and
suppress a complaint from the lava dispatcher that we're using a risky
prompt!)

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3883>

ci: prepare-artifacts: Make the indent here match previously in the file

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3883>

anv: Add pipe_state_for_stage() helper

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3911>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3911>

anv: Use intel_debug_flag_for_shader_stage()

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3911>

spirv: Be consistent when checking for Shader/Kernel

Use == and != instead of the ordered comparisons.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3911>

spirv: Remove outdated SPIR-V decoration warnings

spirv_to_nir warns if it encounters XFB decorations and errors if
it encounters a Stream decoration with value other than 0, despite
the fact that these decorations are in fact handled correctly.

Fixes dEQP-VK.transform_feedback.simple.query_1_*
Fixes: cd4a14be06 "spirv: Handle XFB variable decorations"
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3910>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3910>

nir/builder: Return an integer from nir_get_texture_size

It's convenient in a bunch of cases for nir_get_texture_size to return a
float but it's very unexpected.  This fixes a bug in the R600 NIR code.

Fixes: f718ac62688b "r600/sfn: Add a basic nir shader backend"
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3897>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3897>

nir: Fix the nir_builder include path for nir_builtin_builder

Because it's in double-quotes, it will search the current folder before
any search paths.  Since nir_builder.h and nir_builtin_builder.h are in
the same folder, this guarantees a correct include.  However,
nir/nir_builder.h does not unless the includer's path is set up just
right.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3897>

util: Change os_same_file_description return type from bool to int

This allows communicating that it wasn't possible to determine whether
the two file descriptors reference the same file description. When
that's the case, log a warning in the amdgpu winsys.

In turn, remove the corresponding debugging output from the fallback
os_same_file_description implementation. It depends on the caller if
false negatives are problematic or not.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3879>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3879>

winsys/amdgpu: Make local variable r signed

This is consistent with the return type of the functions whose return
values we assign to it.

No functional change intended.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3879>

nir/lower_ssbo: handle atomics

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2753>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2753>

nir: Add SSBO->global lowering pass

To facilitate lowering SSBOs to globals, we need a load_ssbo_address
intrinsic. This intrinsic takes an SSBO index and loads the address in
global memory of the SSBO (likely implemented via a uniform in the
driver). In the future, we'll support bounds checking, but at the moment
this is not supported (this pass should only be used for trusted
contexts at the moment, i.e. contexts without robustness extensions).

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2753>

panfrost: Rewrite texture descriptor creation logic

Rather than creating partially within the Gallium create function and
monkeypatching on draw time with code split across N different files
with tight Gallium dependencies, let's streamline everything into a
series of maintainable routines in mesa/src/panfrost with no Gallium
dependencies, doing the entire texture creation in one-shot and thus
adding absolutely zero draw-time overhead (since we can allocate a BO
for the descriptor and upload ahead-of-time, so switching textures is as
cheap as switching pointers).

Was this worth it? You know, I'm not sure :|

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3858>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3858>

panfrost: Move format translation to root

Since PIPE formats are now shared across Mesa we can do this, and the
routines themselves are good enough code that I'm happy to move them
here. We'll use them momentarily.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3858>

panfrost: Move pan_afbc.c to root

Now that PIPE formats are shared across Mesa, this well-documented piece
of code is a good fit for root panfrost, let's move it and get a little
closer to taming the mess of resources.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3858>

panfrost: Move checksum routines to root panfrost

These are Gallium-independent and clean code; as is tradition, let's
hoist them up out of the Gallium driver as a bit of yak shaving as we
prepare to untangle the monster that is pan_resource.c

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3858>

util: promote u_debug_memory.c to src/util

When os_memory_debug.h was promoted to src/util, this source-file on
which it depends on when the debug-flag is set on windows was left
out. So let's move this also.

It doesn't seem there's any way of triggering this issue right now, but
it seems better to correct this to avoid this from biting us in the ass
in the future.

Fixes: 88c4680b5a5 ("util: promote u_memory to src/util")
Reviewed-by: Dylan Baker <dylan@pnwbakers>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3844>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3844>

lima: implement PLB PP stream cache

Generating PLB PP stream is expensive. PLB PP stream content depends on
damage, and if damage consists of several rects it's impossible to come
up with a simple key.

Simplify damage to a single bounding box so we have a simple key
and cache PLB PP stream. Cache size is limited to 0.1% of system RAM and
once limit is reached least recently used entries are dropped.

Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3834>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3834>

docs: Update index, relnotes, and release-calendar for 20.0

This includes the release schedule for 20.0.x. Currently there are four
planned releases, but I assume we'll need more before 20.1.0 is ready.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3896>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3896>

Docs: Add 20.0.0 release notes

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3896>

radv: use RADEON_FLAG_ZERO_VRAM when creating the trace BO

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3888>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3888>

radv/winsys: add a new flag that requests zerovram allocations

This introduces RADON_FLAG_ZERO_VRAM.

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/3888>

gallivm: fix crash in emit_get_buffer_size

Seems a bit odd we extract a value from a vector in the first place
(as we always extract the first element), but llvm asserts if using
a zero-vector instead of zero as the index element.
Fixes piglit crashes for example in arb_shader_storage_buffer_object-layout-std140-write-shader.

Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3886>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3886>

gallivm: fix crash with bptc border color sampling

bptc uses fallback for decoding, but still need to handle border color
properly.

v2: adjust piglit gitlab-ci expectations

Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3886>

aco: improve GFX9 1D ddx/ddy assertion

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2547
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3890>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3890>

pan/midgard: Identify clamp(x, -1.0, 1.0) flag

So *that's* what's .unk2 was about :)

We still need to add an opt pass for it, but we can do that further down
the line.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3892>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3892>

panfrost: Remove flush_frontbuffer

A relic from software rasterizers. Hardware drivers generally don't need
to implement this.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3878>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3878>