glapi: remove the final function offset tags

A commit earlier this year reworked out python scripts to use a separate
file for these. Followed by removing support from the parser, and
removing all of the offset tags.

Seems like we either missed a few, or people added them by mistake.
Either way let's nuke the ones that are still around.

Cc: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

winsys/amdgpu/addrlib: do not wrap header inclusion in extern "C"

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

mesa/main: do not wrap header inclusion in extern "C"

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

i915: limit extern "C" hack only for libdrm headers

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

xmesa: do not wrap header inclusion in extern "C"

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

util/sha: do not wrap header inclusion in extern "C"

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

egl/wayland: do not wrap header inclusion in extern "C"

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

gbm: do not wrap header inclusion in extern "C"

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

st/mesa: shader image atoms must be before framebuffer update

The reason is that the shader image atoms call st_finalize_texture, which
may set ST_NEW_FRAMEBUFFER.

This fixes an assertion triggered by a subtest of piglit's
arb_shader_image_load_store-invalid.

v2: add comment explaining order constraints (suggested by Ilia)

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

gallivm: special case TGSI_OPCODE_STORE

This instruction has the resource (buffer or image) as a destination to
represent the writemask for SSBO writes. However, this is obviously not
a "real" destination for the purpose of emitting LLVM IR.

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

tgsi: set correct output mode for RESQ

Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

gallium: add CAPs returning PCI device location

Reviewed-by: Brian Paul <brianp@vmware.com>

winsys/amdgpu: get PCI info

This will be queried by the OpenCL stack using an interop call.

I have tested that the values match lspci.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: set amdgpu metadata before exporting a texture

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: extract the texture descriptor computation into its own function

This will allow this code to be re-used for shader images.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: extract the buffer descriptor computation into its own function

This will allow it to be re-used for shader image descriptors.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: remove resource field from si_sampler_view

view->resource is redundant with view->base.texture, so get rid of it.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: accept pipe_resource in si_sampler_view_add_buffer

and rename .._buffers -> .._buffer

Based loosely on Nicolai's patch. This will make it easier to cherry-pick
Nicolai's patches from his image support branch.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: disable DCC on handle export if expecting write access

This should be okay except that sampler views and images are not re-set.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: add DCC decompression (v2)

This is currently not needed but will be necessary when we have
features that do not work with DCC enabled, such as image stores
and sharing non-scanout surfaces.

v2: Marek: rebase, remove decompression from si_flush_resource (not needed)

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

radeonsi: allocate DCC in the same backing buffer as the texture

To allow sharing textures with DCC enabled.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: disable CMASK on handle export if sharing doesn't allow it (v2)

v2: remove the list of all contexts

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: eliminate fast color clear before sharing

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: don't use fast color clear if sharing doesn't allow it

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: disallow handle export for MSAA & depth textures

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: remember that texture_from_handle was called and its flags

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: check that handle usage doesn't change for a resource

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: disallow reallocation of shared buffers

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: if we can't discard a whole resource, discard the range instead

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: buffer valid range tracking only works with unshared buffers

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: don't set texture metadata for buffers

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: set texture metadata only once

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: clean up r600_texture_get_handle

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: move code initializing texture metadata to its own function

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

winsys/amdgpu: allow drivers to set/get opaque metadata

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: rename winsys buffer_get/set_tiling to buffer_get/set_metadata

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: remove rcs parameter from radeon_winsys::buffer_set_tiling

This was needed for DRM < 2.12.0 where the kernel was rewriting tiling flags
in IBs.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: use a structure for passing tiling flags from/to winsys

and call it radeon_bo_metadata

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium: add external usage flags to resource_from(get)_handle (v2)

This will allow drivers to make better decisions about texture sharing
for DRI2, DRI3, Wayland, and OpenCL.

v2: add read/write flags, take advantage of __DRI_IMAGE_USE_BACKBUFFER

Reviewed-by: Axel Davy <axel.davy@ens.fr>

dri: add backbuffer use flag

This will be used by the next commit.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: dont allow undefined array sizes in ES

This applies the rule to empty declarations.

Fixes:
dEQP-GLES3.functional.shaders.arrays.invalid.empty_declaration_without_var_name_vertex
dEQP-GLES3.functional.shaders.arrays.invalid.empty_declaration_without_var_name_fragment

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

radeon/uvd: increase max height to 4096 for VI and newer

With this issue 'mpv --hwdec=vdpau --vo=vdpau <stream>' fails
for vdpau decode if the stream height is 4096. Vdpau decode of
height upto 4096 is necessary usecase on amdgpu driver for VI
and newer platforms.

The fix is in driver specific implementation of "Decoder
Query Capabilities" API to return 4096 for VI and newer
platforms. With this fix vdpauinfo reports height support as
4096 and mpv for vdpau decode works fine for 4096 height streams.

Signed-off-by: Tamil velan <Tamil-Velan.Jayakumar@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>

winsys/amdgpu: enlarge buffer_indices_hashlist

Enlarge the buffer hashlist to prevent large numbers of misses
due to adding more buffers than can be cached in the hashlist.

The game I tested had CS's with up to 1500 buffers and the overhead
of amdgpu_lookup_buffer for various sizes was:

4096 1.97% (new value)
2048 4.37%
1024 6.92%
512  9.47% (old value)

(percentage of CPU usage in render thread as determined by perf)

The time spent in amdgpu_add_buffer self is ~4.2% in all cases and
for 4096 the time needed to clear the hashlist is still < 0.10%,
so I am not expecting significant regressions.

Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

nvc0: add a new validation path for compute

This makes use of the new state validation interface to be consistent
with 3d.

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

nvc0: rework the validation path for 3D

This exposes an interface for state validation that will be also used
to rework the compute validation path.

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

i965/hsw: Initialize SLM index in state register

For Haswell, we need to initialize the SLM index in the state
register. This can be copied out of the CS header dword 0.

v2:
 * Use UW move to avoid changing upper 16-bits of sr0.1 (mattst88)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94081
Fixes: piglit arb_compute_shader/execution/shared-atomics.shader_test
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "11.2" <mesa-stable@lists.freedesktop.org>
Tested-by: Ilia Mirkin <imirkin@alum.mit.edu> (v1)
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/compute: Skip SIMD8 generation if it can't be used

If the local workgroup size is sufficiently large, then the SIMD8
program can't be used. In this case we can skip generating the SIMD8
program. For complex programs this can save a significant amount of
time.

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

i965/fs: Allow spilling for SIMD16 compute shaders

For fragment shaders, we can always use a SIMD8 program. Therefore, if
we detect spilling with a SIMD16 program, then it is better to skip
generating a SIMD16 program to only rely on a SIMD8 program.

Unfortunately, this doesn't work for compute shaders. For a compute
shader, we may be required to use SIMD16 if the local workgroup size
is bigger than a certain size. For example, on gen7, if the local
workgroup size is larger than 512, then a SIMD16 program is required.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93840
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "11.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: don't always reject shaders with mismatching ifc blocks

Since we store some member qualifiers in the interface type
we need to be more careful about rejecting shaders just because
the pointer doesn't match. Its perfectly valid for some qualifiers
such as precision to not match across shader interfaces.

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

glsl: make interstage_match() static

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

glsl: don't validate ifc blocks using validation meant for variables

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

mesa: Fix error code for GetFramebufferAttachmentParameter in ES 3.0+.

The ES 3.0+ specifications contain the exact same text as the OpenGL
specification, which says that we should return GL_INVALID_OPERATION.

ES 2.0 contains different text saying we should return GL_INVALID_ENUM.

Previously, Mesa chose the error code based on API (GL vs. ES).
This patch makes ES 3.0+ follow the GL behavior.  ES 2 remains as is.

Fixes dEQP-GLES3.functional.fbo.api.attachment_query_empty_fbo.
However, breaks the dEQP-GLES2 variant of the same test for drivers
which silently promote to ES 3.0.  This can be worked around by
exporting MESA_GLES_VERSION_OVERRIDE=2.0, but is a bug in dEQP-GLES2.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Add GL_RED and GL_RG to ES3 effective internal format mapping.

The dEQP-GLES3.functional.fbo.completeness.renderable.texture.
{color0,depth,stencil}.{red,rg}_unsigned_byte tests appear to expect
GL_RED/GL_RG and GL_UNSIGNED_BYTE to map to GL_R8/GL_RG8, rather than
returning an INVALID_OPERATION error.

This makes perfect sense.  However, RED and RG are strangely missing
from the ES 3.0/3.1/3.2 spec's "Effective internal format corresponding
to external format and type" tables.  It may be worth filing a spec bug.

Fixes the 6 dEQP tests mentioned above.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>

nv50,nvc0: make sure to destroy the mutex used for blits

This mutex is initialized when the blitter is created, but it is never
destroyed. This doesn't hurt anything but it makes sense to destroy it
at blitter deletion.

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

gallium/radeon: don't use temporary buffers for persistent mappings

Cc: 11.1 11.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

nir: Add a pass for lower indirect variable dereferences

This new pass lowers load/store_var intrinsics that act on indirect derefs
to if-ladder of direct load/store_var intrinsics.  The if-ladders perform a
simple binary search on the indirect.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

i965/fs/nir: "surface_access::" prefix not needed

"using namespace brw::surface_access" is already present at the
top of the source file.

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

mesa: fix malformed assertion in _image_format_class_to_glenum()

Reviewed-by: Vinson Lee <vlee@freedesktop.org>

program: minor whitespace clean-ups in program_parse_extra.c

st/mesa: conditionally enable GL_NV_vdpau_interop

Only enable it when we compile the state tracker as well.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

radeon/uvd: disable MPEG1

The hardware simply doesn't support that correctly.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>

i965/vec4/nir: no need to use surface_access:: to call emit_untyped_atomic

Now that brw_vec4_visitor::emit_untyped_atomic was removed, there is no need
to explicitly set it.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>

i965/vec4/nir: remove emit_untyped_surface_read and emit_untyped_atomic at brw_vec4_visitor

surface_access emit_untyped_read and emit_untyped_atomic provides the same
functionality.

v2: surface parameter of emit_untyped_atomic is a const, no need to
    specify default predicate on emit_untyped_atomic, use retype
    (Francisco Jerez).

Reviewed-by: Francisco Jerez <currojerez@riseup.net>

i965/vec4: pass the correct src_sz to emit_send at emit_untyped_atomic

If the src is invalid, so src size is zero, the src_sz passed to emit
send should be zero too, instead of a default 1 if we are in a simd4x2
case. This can happens if using emit_untyped_atomic for an atomic
dec/inc.

v2: use the proper src_sz when calling emit_send, instead of just
    avoid loading src at emit_send if BAD_FILE (Francisco Jerez)

Reviewed-by: Francisco Jerez <currojerez@riseup.net>

glcpp: Remove empty mid-rule action which changes test behavior.

Apparently this causes a slight difference in the parser's token
expectations, leading to a different error message.

It seems harmless, but I wanted to be cautious and separate it out.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glcpp: Clean up most empty mid-rule actions left by previous commit.

I didn't want to pollute the previous patch with all the $4 -> $3
changes.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glcpp: Delete unnecessary implicit version resolves.

We now have a bigger hammer.  The HASH_TOKEN NEWLINE rule still needs
to exist to ensure the 146-version-hash-first.c test still passes.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glcpp: Implicitly resolve version after the first non-space/hash token.

We resolved the implicit version directive when processing control lines,
such as #ifdef, to ensure any built-in macros exist.  However, we failed
to resolve it when handling ordinary text.

For example,

        int x = __VERSION__;

should resolve __VERSION__ to 110, but since we never resolved the implicit
version, none of the built-in macros exist, so it was left as is.

This also meant we allowed the following shader to slop through:

        123
        #version 120

Nothing would cause the implicit version to take effect, so when we saw
the #version directive, we thought everything was peachy.

This patch makes the lexer's per-token action resolve the implicit
version on the first non-space/newline/hash token that isn't part of
a #version directive, fulfilling the GLSL language spec:

"The #version directive must occur in a shader before anything else,
 except for comments and white space."

Because we emit #version as HASH_TOKEN then VERSION_TOKEN, we have to
allow HASH_TOKEN to slop through as well, so we don't resolve the
implicit version as soon as we see the # character.  However, this is
fine, because the parser's HASH_TOKEN NEWLINE rule does resolve the
version, disallowing cases like:

        #
        #version 120

This patch also adds the above shaders as new glcpp tests.

Fixes dEQP-GLES2.functional.shaders.preprocessor.predefined_macros.
{gl_es_1_vertex,gl_es_1_fragment}.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

gallium/swr: fix issues preventing a 32-bit build

Not a currently tested configuration, but these couple of small changes
allow a 32-bit build.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94383
Acked-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Brian Paul <brianp@vmware.com>

gallium/swr: remove use of UINT64 from swr_fence

Remove use of a win32-style type leaked from the swr rasterizer.

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

glsl: Add function parameters to the parser symbol table.

In a shader such as:

    struct S { float f; }
    float identity(float S) { return S; }

we would think that "S" in "return S" referred to a structure, even
though it's shadowed by the "float S" parameter in the inner struct.

This led to the parser's grammar seeing TYPE_IDENTIFIER and getting
confused.

Fixes dEQP-GLES2.functional.shaders.scoping.valid.
function_parameter_hides_struct_type_{vertex,fragment}.

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

glsl: Add single declaration variables to the symbol table too.

The lexer/parser use a symbol table to classify identifiers as
variables, functions, or structure types.

For some reason, we neglected to add variables in simple declarations
such as

    int x = 5;

but did add subsequent variables in multi-declarations:

    int x = 5, y = 6; // y gets added, but not x, for some reason

Fixes four dEQP-GLES2.functional.shaders.scoping.valid subcases:
- local_int_variable_hides_struct_type_vertex
- local_int_variable_hides_struct_type_fragment
- local_struct_variable_hides_struct_type_vertex
- local_struct_variable_hides_struct_type_fragment

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

mesa: Change GLboolean to bool in GenerateMipmap target checker.

This is not API facing, so just use bool.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa: Make GenerateMipmap check the target before finding an object.

If glGenerateMipmap was called with a bogus target, then it would
pass that to _mesa_get_current_tex_object(), which would raise a
_mesa_problem() telling people to file bugs.  We'd then do the
proper error checking, raise an error, and bail.

Doing the check first avoids the _mesa_problem().  The DSA variant
doesn't take a target parameter, so we leave the target validation
exactly as it was in that case.

Fixes one dEQP GLES2 test:
dEQP-GLES2.functional.negative_api.texture.generatemipmap.invalid_target.

v2: Rebase on Antia's recent patch to this area.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com> [v1]
Reviewed-by: Matt Turner <mattst88@gmail.com>

gm107/ir: add emission for ATOMS

This allows to perform atomic operations on shared memory.

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

tgsi: fix parsing of shared memory declarations

The SHARED TGSI keyword is only allowed with TGSI_FILE_MEMORY and not
with TGSI_FILE_BUFFER. I have found this by using the nouveau_compiler
from command line.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.2" <mesa-stable@lists.freedesktop.org>

gm107/ir: add emission for BAR

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

gk110/ir: add missing src predicate emission for BAR.RED

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

gk110/ir: allow to emit immediates for BAR

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

gk110/ir: fix wrong emission of BAR.SYNC

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

nvc0/ir: make sure that thread count immediate for BAR fit

The limit of the thread count immediate value is 12 bits.

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

svga: add new surface-write-flushes HUD query

To know when we're flushing the command buffer because we need to
write to surface in the command buffer.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

svga: add new flush-time HUD query

To measure the time spent flushing the command buffer.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

svga: also dump SVGA3D_BUFFER surfaces in svga_screen_cache_dump()

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

mesa: flip current tf object back to default if current is being deleted

In the rather unusual case of Bind + Delete, we need to make sure that
we unbind the current tf object.

Fixes dEQP-GLES3.functional.lifetime.delete_bound.transform_feedback

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: avoid stack smashing when there are too many attributes

This fixes a crash in

dEQP-GLES3.functional.transform_feedback.array_element.separate.points.lowp_mat3x2

and likely others. The vertex shader has > 16 input variables (without
explicit locations), which causes us to index outside of the to_assign
array.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>

i965/vec4: Propagate swizzles correctly during copy propagation.

This simplifies the code that iterates over the per-component values
found in the matching copy_entry struct and checks whether the
register regions that were copied to each component are similar enough
to be treated as a single (reswizzled) value which can be propagated
into the current instruction.

Aside from being scattered between opt_copy_propagation(),
try_copy_propagate(), and try_constant_propagate(), what I found
terribly confusing about the preexisting logic was that
opt_copy_propagation() tried to reorder the array of values according
to the swizzle of the instruction source, which meant one would have
had to invert the reordering applied at the top level in order to find
out which component to take from each value (we were just taking the
i-th component from the i-th value, which is not correct in general).
The saturate mask was also being swizzled incorrectly.

This consolidates the logic for matching multiple components of a
copy_entry into a single function which returns the result as a
regular src_reg on success, as if the copy had been performed with a
single MOV instruction copying all components of the src_reg into the
destination.

Fixes several ARB_vertex_program MOV test-cases from:
 https://cgit.freedesktop.org/~kwg/piglit/log/?h=arb_program

Acked-by: Matt Turner <mattst88@gmail.com>

i965: Don't try copy propagation if constant propagation succeeded.

It cannot get any better.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/vec4: Use swizzle() to swizzle immediates during constant propagation.

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

i965: Add support for swizzling arbitrary immediates to (brw_)swizzle().

Scalar immediates used to be handled correctly by swizzle() (as the
identity) but since commit 58fa9d47b536403c4e3ca5d6a2495691338388fd it
will corrupt the contents of the immediate.  Vector immediates were
never handled correctly, but we had ad-hoc code to swizzle VF
immediates in the vec4 copy propagation pass.  This takes care of
swizzling V and UV in addition.

v2: Don't implement swizzling of V/UV immediates (Matt).  If you need
    to swizzle an integer vector immediate in the future apply the
    following diff to go back to v1:

--- a/src/mesa/drivers/dri/i965/brw_eu.c
+++ b/src/mesa/drivers/dri/i965/brw_eu.c
@@ -119,11 +119,10 @@ brw_swap_cmod(uint32_t cmod)
 static unsigned
 imm_shift(enum brw_reg_type type, unsigned i)
 {
-   assert(type != BRW_REGISTER_TYPE_UV && type != BRW_REGISTER_TYPE_V &&
-          "Not implemented.");
-
    if (type == BRW_REGISTER_TYPE_VF)
       return 8 * (i & 3);
+   else if (type == BRW_REGISTER_TYPE_UV || type == BRW_REGISTER_TYPE_V)
+      return 4 * (i & 7);
    else
       return 0;
 }

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

i965: Pass symbolic swizzle to brw_swizzle() as a single argument.

And replace brw_swizzle1() with brw_swizzle().  Seems slightly cleaner
and will allow reusing brw_swizzle() in the vec4 back-end more easily.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

nvc0: reset TFB bufctx when we no longer hold a reference to the buffers

This fixes some use-after-free situations in dEQP when an xfb state is
removed, and then a clear is triggered, which only does a partial
validation. It would attempt to read the no-longer-valid buffers,
resulting in crashes.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>

nv50/ir: using sampleid/pos shouldn't force per-sample interpolation

See https://www.khronos.org/bugzilla/show_bug.cgi?id=1462

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

st/mesa: don't force per-sample interp if only sampleid/pos are used

The OES extensions clarify this behaviour to differentiate between
per-sample invocation and per-sample interpolation. Using sampleid/pos
will force per-sample invocation but not per-sample interpolation.

See https://www.khronos.org/bugzilla/show_bug.cgi?id=1462

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

swrast: fix GL_ANY_SAMPLES_PASSED values in Result

Since commit 922be4eab, the expectation is that the query result
contains the correct value. Unfortunately swrast does not distinguish
between GL_SAMPLES_PASSED and GL_ANY_SAMPLES_PASSED. As a result, we
must fix up the query result in a post-draw fixup.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94274
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Tested-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "11.2" <mesa-stable@lists.freedesktop.org>

st/mesa: 78-column wrapping in st_extensions.c

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

gallium/util: add new comments, assertions in u_debug_refcnt.c

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

gallium/util: update comments and URL in u_debug_refcnt.c

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

gallium/util: make stream variable static in u_debug_refcnt.c

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

gallium/util: re-indent u_debug_refcnt.[ch]

Wrap comments to 78 columns, etc.

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>