nvc0: dump program binary only when NV50_PROG_DEBUG is set

When the chipset is forced with NV50_PROG_CHIPSET, we actually
only want to output the binary if NV50_PROG_DEBUG is also
enabled. Otherwise, this pollutes the shader-db output.

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

nir: Fix the control flow tests for nir_loop_first_block changes

Commit 2ed17d46de045404042f13c6591895a1cf31b167 changed
nir_loop_first_cf_node and friends to return a nir_block instead of a
nir_cf_node.  This broke one of the NIR control flow tests.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98128

docs: mark ARB_compute_variable_group_size as done for nvc0

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

nvc0: expose ARB_compute_variable_group_size

Only expose 512 threads/block on Fermi to not be limited by
32 GPRs/thread.

v4: - use 512 threads on Fermi, 1024 on Kepler+

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

nv50/ir: set number of threads/block for variable local size

When a variable local size is defined as specified by
ARB_compute_variable_group_size, the fixed local size is set to 0
and a SIGFPE occurs when we compute the maximum number of regs.

This allows to use 64 GPRs/thread.

v4: - use 512 threads on Fermi, 1024 on Kepler+

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

st/mesa: expose ARB_compute_variable_group_size

This extension is only exposed if the underlying driver supports
ARB_compute_shader and if PIPE_COMPUTE_MAX_VARIABLE_THREADS_PER_BLOCK
is set.

v3: - initialize max_variable_threads_per_block to 0
v2: - expose the ext based on that new cap

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: add support for dispatching a variable local size

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: add mapping for SYSTEM_VALUE_LOCAL_GROUP_SIZE

gl_LocalGroupSizeARB can be translated into TGSI_SEMANTIC_BLOCK_SIZE
which represents the block size in threads.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium: add PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK

v3: - use a new case statement in r600_pipe_common.c
    - fix compilation of softpipe...

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

glsl: add gl_LocalGroupSizeARB as a system value

v2: - only add it if the ext is enabled (Ilia)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

glsl/linker: handle errors when a variable local size is used

Compute shaders can now include a fixed local size as defined by
ARB_compute_shader or a variable size as defined by
ARB_compute_variable_group_size.

v2: - update formatting spec quotations (Ian)
    - various cosmetic changes (Ian)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

glsl: reject compute shaders with fixed and variable local size

The ARB_compute_variable_group_size specification explains that
when a compute shader includes both a fixed and a variable local
size, a compile-time error occurs.

v2: - update formatting spec quotations (Ian)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

glsl: process local_size_variable input qualifier

This is the new layout qualifier introduced by
ARB_compute_variable_group_size which allows to use a variable work
group size.

v4: - add missing '%s' in the monster format string

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

glsl: add enable flags for ARB_compute_variable_group_size

This also initializes the default values for the standalone compiler.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

mesa/main: add support for ARB_compute_variable_groups_size

v5: - replace fixed_local_size by !LocalSizeVariable (Nicolai)
v4: - slightly indent spec quotes (Nicolai)
    - drop useless _mesa_has_compute_shaders() check (Nicolai)
    - move the fixed local size outside of the loop (Nicolai)
    - add missing check for invalid use of work group count
v2: - update formatting spec quotations (Ian)
    - move the total_invocations check outside of the loop (Ian)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

glapi: add entry points for GL_ARB_compute_variable_group_size

v2: - correctly sort that new extension (Ian)
    - fix up the comment (Ian)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

nv50/ir: optimize sub(a, 0) to a

helped some ue4 demos and divinity OS shaders

total instructions in shared programs : 2818674 -> 2818606 (-0.00%)
total gprs used in shared programs    : 379273 -> 379273 (0.00%)
total local used in shared programs   : 9505 -> 9505 (0.00%)
total bytes used in shared programs   : 25837792 -> 25837192 (-0.00%)

                local        gpr       inst      bytes
    helped           0           0          33          33
      hurt           0           0           0           0

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>

st/mesa: move all sampler view code into new st_sampler_view.[ch] files

Previously, the sampler view code was scattered across several different
files.

Note, the previous REALLOC(), FREE() for st_texture_object::sampler_views
are replaced by realloc(), free() to avoid conflicting macros in Mesa vs.
Gallium.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Acked-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: optimize pipe_sampler_view validation

Before, st_get_texture_sampler_view_from_stobj() did a lot of work to
check if the texture parameters matched the sampler view (format,
swizzle, min/max lod, first/last layer, etc).  We did this every time
we validated the texture state.

Now, we use a ctx->Driver.TexParameter() callback and a couple other
checks to proactively release texture views when we know that
view-related parameters have changed.  Then, the validation step is
simplified:
- Search the texture's list of sampler views (just match the context).
- If found, we're done.
- Else, create a new sampler view.

There will never be old, out-of-date sampler views attached to texture
objects that we have to test.

Most apps create textures and set the texture parameters once.  This
make sampler view validation much cheaper for that case.

Note that the old texture/sampler comparison code has been converted
into a set of assertions to verify that the sampler view is in fact
consistent with the texture parameters.  This should help to spot any
potential regressions.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: call ctx->Driver.TexParameter() in texture_buffer_range()

To inform drivers of texture buffer offset/size changes, as we do for
other texture object parameters.

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

st/mesa: consolidate view format setup code

Before, we had code to compute the sampler view's format spread across two
different functions: in update_single_texture() and
st_get_texture_sampler_view_from_stobj().  Now it's all in one new function.

Also, use _mesa_texture_base_format() to simplify the code.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: add some const qualifiers in st_atom_texture.c

And minor code reformatting.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: simplify some code in get_texture_format_swizzle()

There's no need to cast to st_texture_image.  Just use gl_texture_image.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: make _mesa_texture_buffer_range() static

Not called from any other file.  Also, add a comment.

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

mesa: add const qualifier, comment on can_avoid_reallocation()

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

mesa: add comment/assertion on get_tex_level_parameter_buffer()

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

nir: Remove some no longer needed asserts

Now that the NIR casting functions have type assertions, we have a bunch of
assertions that aren't needed anymore.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Make nir_foo_first/last_cf_node return a block instead

One of NIR's invariants is that control flow lists always start and end
with blocks.  There's no good reason why we should return a cf_node from
these functions since we know that it's always a block.  Making it a block
lets us remove a bunch of code.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Add asserts to the casting functions

This makes calling nir_foo_as_bar a bit safer because we're no longer 100%
trusting in the caller to ensure that it's safe.  The caller still needs to
do the right thing but this ensures that we catch invalid casts with an
assert rather than by reading garbage data.  The one downside is that we do
use the casts a bit in nir_validate and it's not a validate_assert.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

gallium/hud: Remove superfluous debug

No longer required.

Signed-off-by: Steven Toth <stoth@kernellabs.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

amd: add amd_kernel_code_t.h to the sources list

Otherwise it won't be picked in the tarball and the build will fail.

Fixes: 91ec6e56649 ("radeonsi/compute: Use the HSA abi for non-TGSI
compute shaders v3")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

svga: add svga_mksstats.h to the sources list

Otherwise it won't be picked in the tarball and the build will fail.

Fixes: 0035f7f1365 ("svga: add guest statistic gathering interface")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

glx: rename choose_visual(), drop const argument

The function deals with fb (style) configs, thus using "visual"
in the name is misleading. Which in itself had led to the use of
fbconfig_style_tags argument.

Rename the function to reflect what it does and drop the unneeded
argument.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glx: return GL_FALSE from glx_screen_init where applicable.

Return GL_FALSE if we fail to find any fb/visual configs, otherwise we
end up with all sorts of chaos further down the GLX stack.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glx: correctly mask the drawableType for GLX_ARB_fbconfig_float

The comment/spec says - only for pbuffer drawables, while the code
clears the window/pixmap bit. Practise what you preach and apply the
trivial tweak. In practise this should not cause functional change.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

autoconf: Make header install distinct for various APIs (v2)

This fixes a problem where GL headers would only get installed if
glx was enabled.  So if osmesa was enabled but not glx, then the
GL headers required by osmesa would be missing from the install.

v2: Dropped unneeded mesa_glinterop.h redundant osmesa.h install

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Chuck Atkins <chuck.atkins@kitware.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

mesa: annotate AttribFuncsARB[] as const

It's read-only data, so annotate it accordingly.

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

mapi/glapi: remove unused _glapi_check_table()

Similar to earlier commit - symbol was never part of the public API so
we're safe to remove it.

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

glapi/hgl: remove the final user of _glapi_check_table()

The symbol is a no-op since, the EXTRA_DEBUG macro is not set in the
build. Unused by !Haiku people/platforms since 2010 (commit
a73c6540d9a7f6e26d8568ba2fc522cb865f0a6c) while the Haiku C++ wrapper
has no obvious users.

Cc: Alexander von Gluck IV <kallisti5@unixzen.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

mapi/glapi: remove unused _glapi_check_table_not_null

Function was never part of the API/ABI and the final user was removed
with commit a73c6540d9a7f6e26d8568ba2fc522cb865f0a6c, back in 2010.

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

st/xvmc/tests: force enable assertions

Similar to the other 'tests', enable assertions in xvmc_bench.

This silences the GCC warnings about unused-variable(s), makes the
program actually useful, as the XvMC API called. Atm the function
calls are omitted, since they're called within the assert.

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

anv: automake: ship intel_icd.json.in in the tarball

Otherwise we'll fail to (re)generate intel_icd.json.

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

intel: automake: reference the correct header

The header was renamed with earlier commit, so update the
Makefile.sources respectively.

{vulkan/genX_multisample.h => common/gen_sample_positions.h}

Fixes: c779ad3e661("intel: Move Vulkan sample positions to common code")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

intel: aubinator: add missing return characters

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

nir: Delete open coded type printing.

glsl_print_type() prints arrays of arrays incorrectly.  For example,
a type with name float[3][7] would be printed as float[7][3].  (This
is an array of length 3 containing arrays of 7 floats.)  cdecl says
that the type name is correct.

glsl_print_type() doesn't really do anything above and beyond printing
type->name, and glsl_print_struct() wasn't used at all.  So, drop them.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

anv: fix GetPhysicalDeviceProperties to return timestampPeriod in ns

According to chapters 16.5. (Timestamp Queries) and 30.2 (Limits) of the
Vulkan Specification 1.0.29, the .limits.timestampPeriod field returned
by vkGetPhysicalDeviceProperties is measured in nanoseconds, not in
seconds.

Signed-off-by: Philipp Zabel <philipp.zabel@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: remove remaining tabs in brw_draw.c

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

i965: get inputs read from nir info

This is a step towards dropping the GLSL IR version of
do_set_program_inouts() in i965 and moving towards native nir support.

This is important because we want to eventually convert to nir and
use its optimisations passes before we can call this GLSL IR pass.

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

i965: get outputs written from nir info

This is a step towards dropping the GLSL IR version of
do_set_program_inouts() in i965 and moving towards native nir support.

This is important because we want to eventually convert to nir and
use its optimisations passes before we can call this GLSL IR pass.

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

i965: get outputs read from nir info

This is a step towards dropping the GLSL IR version of
do_set_program_inouts() in i965 and moving towards native nir support.

This is important because we want to eventually convert to nir and
use its optimisations passes before we can call this GLSL IR pass.

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

i965: remove remaining tabs in brw_wm.c

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

mesa: remove the UsesDFdy flag

Seems the last user of this was removed in 08bc74e69.

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

i965: get uses discard from nir info

This is a step towards dropping the GLSL IR version of
do_set_program_inouts() in i965 and moving towards native nir support.

This is important because we want to eventually convert to nir and
use its optimisations passes before we can call this GLSL IR pass.

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

i965: get uses texture gather from nir info

This is a step towards dropping the GLSL IR version of
do_set_program_inouts() in i965 and moving towards native nir support.

This is important because we want to eventually convert to nir and
use its optimisations passes before we can call this GLSL IR pass.

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

i965: Eliminate brw->cs.prog_data pointer.

Just say no to:

-   brw->cs.base.prog_data = &brw->cs.prog_data->base.base;

We'll just use the brw_stage_prog_data pointer in brw_stage_state
and downcast it to brw_cs_prog_data as needed.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>

i965: Eliminate brw->wm.prog_data pointer.

Just say no to:

-   brw->wm.base.prog_data = &brw->wm.prog_data->base.base;

We'll just use the brw_stage_prog_data pointer in brw_stage_state
and downcast it to brw_wm_prog_data as needed.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>

i965: Eliminate brw->gs.prog_data pointer.

Just say no to:

-   brw->gs.base.prog_data = &brw->gs.prog_data->base.base;

We'll just use the brw_stage_prog_data pointer in brw_stage_state
and downcast it to brw_gs_prog_data as needed.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>

i965: Eliminate brw->tes.prog_data pointer.

Just say no to:

-   brw->tes.base.prog_data = &brw->tes.prog_data->base.base;

We'll just use the brw_stage_prog_data pointer in brw_stage_state
and downcast it to brw_tes_prog_data as needed.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>

i965: Eliminate brw->tcs.prog_data pointer.

Just say no to:

-   brw->tcs.base.prog_data = &brw->tcs.prog_data->base.base;

We'll just use the brw_stage_prog_data pointer in brw_stage_state
and downcast it to brw_tcs_prog_data as needed.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>

i965: Eliminate brw->vs.prog_data pointer.

Just say no to:

-   brw->vs.base.prog_data = &brw->vs.prog_data->base.base;

We'll just use the brw_stage_prog_data pointer in brw_stage_state
and downcast it to brw_vs_prog_data as needed.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>

i965: Introduce downcast helpers for prog_data structures.

Similar to brw_context(...), intel_texture_object(...), and so on.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>

i965/sync: Rename awkward variable

What is the difference between a 'driver_fence' and a 'fence'? Do the
characters 'driver_' add anything helpful? Nope. They do, though, add an
extra 7 chars and pull your eyeballs away to ask "huh? what's that?" one
microsecond too many.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/sync: Rename intel_syncobj.c -> brw_sync.c

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/sync: Replace 'intel' prefix with 'brw'

This is yet another patch for the great renaming begun long ago.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/sync: Fix uninitalized usage and leak of mutex

We locked an unitialized mutex in the callstack
    glClientWaitSync
    intel_gl_client_wait_sync
    brw_fence_client_wait_sync
because we forgot to initialize it in intel_gl_fence_sync.
(The EGLSync codepath didn't have this bug. It initialized the mutex in
intel_dri_create_sync).

We also forgot to tear down (mtx_destroy) the mutex when destroying
the sync object.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

nir: Use the correct infos structure for copying atomic sources

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Cc: "12.0" <mesa-dev@lists.freedestkop.org>

nvc0: dump program binary when chipset has been forced

Currently, program binaries are only dumped at upload time, but
when the chipset has been forced via NV50_PROG_CHIPSET we might
want to show the generated code, especially with shaderdb.

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

radeonsi: fix texture border colors for compute shaders

There are VM faults without this.

Cc: 12.0 <mesa-stable@lists.freedesktop.org>
Acked-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon/winsyses: set reasonable max_alloc_size

which is returned for GL_MAX_TEXTURE_BUFFER_SIZE.
It doesn't have any other use at the moment.
Bigger allocations are not rejected.

This fixes GL45-CTS.texture_buffer.texture_buffer_max_size on Bonaire.

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

radeonsi: fix interpolateAt opcodes for .zw components

Not returning garbage in .zw seems pretty important.

This fixes:
GL45-CTS.shader_multisample_interpolation.render.interpolate_at_*_check.*

Cc: 11.2 12.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: add assertions to validate interpolation flags

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

radeonsi: interpolate colors after interpolation weight shuffling

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

tgsi/scan: don't set interp flags for inputs only used by INTERP (v2)

(v1 pushed, then reverted)

This fixes 9 randomly failing tests on radeonsi:
  GL45-CTS.shader_multisample_interpolation.render.interpolate_at_centroid.*

v2: use input_interpolate[input] (correct) instead of
    input_interpolate[index] (incorrect)

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

ddebug: dump most driver information with GALLIUM_DDEBUG=always

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

nv50/ra: let simplify return an error and handle that

fixes a crash in the case simplify reports an error

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

intel/blorp: Use documented RECTLIST vertex positions

Use the vertex positions described in the PRMs. This has no effect on
rendering but quiets the simulator warnings seen when the vertices
appear out of order.

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>

anv/meta: Roll clear_image into CmdClearDepthStencilImage

It is now the only caller so there's no sense in keeping things split out.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv: Use blorp for VkCmdFillBuffer

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

egl: Implement EGL_KHR_debug (v2)

Wire up the debug entrypoints to EGL dispatch, and add the extension
string to the client extension list.

v2:
- Lots of style fixes
- Fix missing EGLAPIENTRYs
- Factor out valid attribute check
- Lock display in eglLabelObjectKHR as needed, and use RETURN_EGL_*
- Move "EGL_KHR_debug" into asciibetical order in client extension
  string

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Emil Velikov <emil.veliko@collabora.com>

egl: Track EGL_KHR_debug state when going through EGL API calls (v3)

This decorates every EGL entrypoint with _EGL_FUNC_START, which records
the function name and primary dispatch object label in the current
thread state. It also adds debug report functions and calls them when
appropriate.

This would be useful enough for debugging on its own, if the user set a
breakpoint when the report function was called. We will also need this
state tracked in order to expose EGL_KHR_debug.

v2:
- Clear the object label in more cases in _eglSetFuncName
- Pass draw surface (if any) to _EGL_FUNC_START in eglSwapInterval

v3:
- Set dummy thread's CurrentAPI to EGL_OPENGL_ES_API not zero
- Less ?: in _eglSetFuncName

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Emil Velikov <emil.veliko@collabora.com>

intel: aubinator: pack supported generations into an array

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965/l3: Add explicit way size calculation for bxt

There should be no functional change here because Broxton and CHV are
both gt1. Without this code however, it might seem like broxton support
is missing.

While here, put the gt1 check in front to hopefully short-circuit the
condition for the mobile cases.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>

virgl: Fix build regression of commit 8a943564

st/mesa: enable GL_KHR_robustness

The difference to the virtually identical ARB_robustness (which is already
enabled unconditionally) is miniscule and handled elsewhere, but this cap
seems like the right thing to require for this extension.

v2: drop the device reset cap requirement (Ilia)

Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>

gallium/radeon: implement set_device_reset_callback

Check for device reset on flush. It would be nicer if the kernel just
reported this as an error on the submit ioctl (and similarly for fences),
but this will do for now.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: set a device reset callback when available

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: extract conversion from pipe_reset_status to GLenum

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

ddebug: add pass-through of set_device_reset_callback

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

gallium: add pipe_context::set_device_reset_callback

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

virgl: use the new parent/child pools for transfers

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

vc4: use the new parent/child pools for transfers

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

freedreno: use the new parent/child pools for transfers

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

r300: use the new parent/child pools for transfers (v2)

v2: slab_alloc_st -> slab_alloc

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

gallium/radeon: use the new parent/child pools for transfers

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97894
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

util/slab: re-design to allow migration between pools (v3)

This is basically a re-write of the slab allocator into a design where
multiple child pools are linked to a parent pool. The intention is that
every (GL, pipe) context has its own child pool, while the corresponding
parent pool is held by the winsys or screen, or possibly the GL share group.

The fast path is still used when objects are freed by the same child pool
that allocated them. However, it is now also possible to free an object in a
different pool, as long as they belong to the same parent. Objects also
survive the destruction of the (child) pool from which they were allocated.

The slow path will return freed objects to the child pool from which they
were originally allocated. If that child pool was destroyed, the corresponding
page is considered an orphan and will be freed once all objects in it have
been freed.

This allocation pattern is required for pipe_transfers that correspond to
(GL) buffer object mappings when the mapping is created in one context
which is later destroyed while other contexts of the same share group live
on -- see the bug report referenced below.

Note that individual drivers do need to migrate to the new interface in
order to benefit and fix the bug.

v2: use singly-linked lists everywhere
v3: use p_atomic_set for page->u.num_remaining

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97894

util: use GCC atomic intrinsics with explicit memory model

This is motivated by the fact that p_atomic_read and p_atomic_set may
somewhat surprisingly not do the right thing in the old version: while
stores and loads are de facto atomic at least on x86, the compiler may
apply re-ordering and speculation quite liberally. Basically, the old
version uses the "relaxed" memory ordering.

The new ordering always uses acquire/release ordering. This is the
strongest possible memory ordering that doesn't require additional
fence instructions on x86. (And the only stronger ordering is
"sequentially consistent", which is usually more than you need anyway.)

I would feel more comfortable if p_atomic_set/read in the old
implementation were at least using volatile loads and stores, but I
don't see a way to get there without typeof (which we cannot use here
since the code is compiled with -std=c99).

Eventually, we should really just move to something that is based on
the atomics in C11 / C++11.

Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

i965: use L3 data cache for SSBOs

Anv programs the hardware to use L3 data cache if we use either SSBOs or
images in the shaders, we can program i965 the same way.

gl_shader_program has a bit of a confusing named field with
'NumAtomicBuffers'. It doesn't tell how many buffers are accessed by the
shader in an atomic way but instead the number of atomic counters
manipulated by the shader.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>

mesa: Raise INVALID_ENUM in FramebufferTexture*D for unknown textargets.

ES3-CTS.functional.negative_api.buffer.framebuffer_texture2d expects
glFramebufferTexture[123]D to raise GL_INVALID_ENUM when
supplied a completely bogus textarget parameter (i.e. 0xffffffff).

This is at odds with the spec.  GLES 3.1 says:

   "An INVALID_OPERATION error is generated if texture is not zero and
    textarget is not one of TEXTURE_2D, TEXTURE_2D_MULTISAMPLE, or one
    of the cube map face targets from table 8.21."

(and GLES 3.0 and GL 4.5 both have similar text).  However, GL has a
general guideline that says:

   "If a command that requires an enumerated value is passed a symbolic
    constant that is not one of those specified as allowable for that
    command, an INVALID_ENUM error is generated."

Apparently other vendors reconcile these two rules as follows: GL should
raise INVALID_OPERATION for actual texture target enumeration values
which are not allowed for this particular glFramebufferTexture*D call.
Any value that is not a texture target should result in GL_INVALID_ENUM.

For example, glFramebufferTexture2D with GL_TEXTURE_1D would result in
INVALID_OPERATION because it is a real texture target, but not allowed
for the 2D version of the function.  But calling it with GL_FRONT would
result in INVALID_ENUM, as that isn't even a texture target.

Fixes:
- {ES3-CTS,dEQP-GLES3}.functional.negative_api.buffer.framebuffer_texture2d
- {ES31-CTS,ES32-CTS,dEQP-GLES31}.functional.debug.negative_coverage.get_error.buffer.framebuffer_texture2d

References: https://gitlab.khronos.org/opengl/cts/merge_requests/387
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

mesa: Reorganize check_textarget().

Having one top-level switch statement covering all known texture targets
will make the next change easier to implement.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

aubinator: use the correct format specifier for printing ptrdiff_t.

Fixes more warnings in 32-bit builds.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>