nir: Add a array-of-vector variable shrinking pass

This pass looks for variables with vector or array-of-vector types and
narrows the type to only the components used.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

intel/nir: Use the new structure and array splitting passes

We call structure splitting once because it is guaranteed to split all
the structures in the entire shader in one go.  We call array splitting
in the loop in case future optimizations turn indirects into direct
dereferences and we can split more arrays.

Shader-db results on Kaby Lake:

    total instructions in shared programs: 15177605 -> 15177605 (0.00%)
    instructions in affected programs: 0 -> 0
    helped: 0
    HURT: 0

This is unsurprising because nir_lower_vars_to_ssa already effectively
does structure and array splitting internally.  It doesn't actually
split the variables but it's ability to reason about aliasing in the
presence of arrays and structures and pick out scalars or vectors to be
lowered to SSA values is fairly advanced.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

nir: Add an array splitting pass

This pass looks for array variables where at least one level of the
array is never indirected and splits it into multiple smaller variables.

This pass doesn't really do much now because nir_lower_vars_to_ssa can
already see through arrays of arrays and can detect indirects on just
one level or even see that arr[i][0][5] does not alias arr[i][1][j].
This pass exists to help other passes more easily see through arrays of
arrays.  If a back-end does implement arrays using scratch or indirects
on registers, having more smaller arrays is likely to have better memory
efficiency.

v2 (Jason Ekstrand):
 - Better comments and naming (some from Caio)
 - Rework to use one hash map instead of two

v2.1 (Jason Ekstrand):
 - Fix a couple of bugs that were added in the rework including one
   which basically prevented it from running

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

nir: Add a structure splitting pass

This pass doesn't really do much now because nir_lower_vars_to_ssa can
already see through structures and considers them to be "split".  This
pass exists to help other passes more easily see through structure
variables.  If a back-end does implement arrays using scratch or
indirects on registers, having more smaller arrays is likely to have
better memory efficiency.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

nir/types: Add array_or_matrix helpers

Reviewed-by: Thomas Helland<thomashelland90@gmail.com>

i965: don't include compute resources in "Combined" limits

The combined limits should only include shader stages that can be active
at the same time.  We don't need to include compute.

See also cff290df4c09547cd2cb3b129ec59bdebdadba90 for st/mesa.

Unbreaks i965 from assert failing on driver load since Marek's
45f87a48f94148b484961f18a4f1ccf86f066b1c, which dropped the core
Mesa capabilities before adjusting driver limits down to match.

radeonsi: increase the maximum UBO size to 2 GB

Same as the closed driver.

This causes a failure in GL45-CTS.compute_shader.max, which has a trivial
bug.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

radeonsi: bump MAX_GS_INVOCATIONS

same as the closed driver

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

gallium: add PIPE_CAP_MAX_SHADER_BUFFER_SIZE

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

gallium: add PIPE_CAP_MAX_GS_INVOCATIONS

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

tgsi/ureg: don't call tgsi_sanity when it's too slow

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

st/mesa: fix up uniform limits to be able to expose large UBOs

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

st/mesa: don't include compute resources in "Combined" limits

The combined limits should only include shader stages that can be active
at the same time.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

st/mesa: set ctx->Const.SubPixelBits

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

glsl: fix error checking against MAX_UNIFORM_LOCATIONS

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

mesa: make MaxCombinedUniformComponents 64-bit to allow large UBOs

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

mesa: add ctx->Const.MaxGeometryShaderInvocations

radeonsi wants to report a different value

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

mesa: don't include compute resources in MAX_COMBINED_* limits

5 is the maximum number of shader stages that can be used by 1 execution
call at the same time (e.g. a draw call). The limit ensures that each
stage can use all of its binding points.

Compute is separate and doesn't need the 5x multiplier.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

mesa: bump GL_MAX_ELEMENTS_INDICES and GL_MAX_ELEMENTS_VERTICES

same number as our closed GL driver

v2: don't use MaxArrayLockSize

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

mesa: remove incorrect change for EXT_disjoint_timer_query

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

glapi: actually implement GL_EXT_robustness for GLES

The extension was exposed but not the functions.

This fixes:
    dEQP-GLES31.functional.debug.negative_coverage.get_error.buffer.readn_pixels
    dEQP-GLES31.functional.debug.negative_coverage.get_error.state.get_nuniformfv
    dEQP-GLES31.functional.debug.negative_coverage.get_error.state.get_nuniformiv

Cc: 18.1 18.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

intel/decoder: Decode SFIXED values.

This lets us example SAMPLER_STATE's LOD Bias field, among other things.

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

travis: use python3 for the autoconf builds

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

configure: allow building with python3

Pretty much all of the scripts are python2+3 compatible.
Check and allow using python3, while adjusting the PYTHON2 refs.

Note:
 - python3.4 is used as it's the earliest supported version
 - python3 chosen prior to python2

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

bin/git_sha1_gen.py: remove execute bit/shebang

The script is executed explicitly via the build system, that uses
PYTHON/prog_python and equivalent.

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

vk/wsi: avoid reading uninitialised memory

It will be ignored by x11_swapchain_result() anyway (because reaching
the `fail` label without setting `result` means the swapchain status was
already a hard error), but the compiler still complains about reading
uninitialised memory.

While at it, drop the unused assignment right before returning.

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

egl: drop unused _EGL_BUILT_IN_DRIVER_DRI2

Unused since b174a1ae720cb404738c "egl: Simplify the "driver" interface".

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>

radv/gfx9: implement coherent shaders for VK_ACCESS_SHADER_READ_BIT

Single-sample color and single-sample depth (not stencil)
are coherent with shaders.

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

bin/install_megadrivers.py: Remove shebang and executable bit

Since the script is never executed directly, but launched by Meson as an
argument to the Python interpreter, those are not needed any more.

In addition, they are the reason this script was missed when I moved the
Meson buildsystem to Python 3, so removing them helps avoiding future
confusion.

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

meson: Run the install script with Python 3

The script was being run directly as an executable, and it has a
Python 2 shebang.

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

glsl: remove execute bit and shebang from python tests

Just like the rest of the tree - these should be run either as part of
the build system check target, or at the very least with an explicitly
versioned python executable.

Fixes: db8cd8e3677 ("glcpp/tests: Convert shell scripts to a python script")
Fixes: 97c28cb0823 ("glsl/tests: Convert optimization-test.sh to pure python")
Fixes: 3b52d292273 ("glsl/tests: reimplement warnings-test in python")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>

docs: update required mako version

The requirement was bumped a while back, but we forgot to update the
docs.

Fixes: ed871af91c2 ("configure.ac: raise Mako required version to
0.8.0")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>

configure: use distutils in ax_check_python_mako_module

Handling the version comparison by hand is a bad idea. Python has a handy
module distutils for that - use it.

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

configure: enforce python 2.7 with AM_PATH_PYTHON

Currently we use AC_CHECK_PROGS looking for python2.7, python2 and
finally python. That is due to the varying names used across the
different OS.

Use the handy AM_PATH_PYTHON which finds the correct name and checks for
the version.

Note: python2.7 has been an unofficial requirement for quite some time.
Update the docs to reflect that.

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

i965: Enable INTEL_shader_atomic_float_minmax on Gen9+

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

i965: Sort Gen9+ extension enables

This is a strictly alphabetic sort, as is done in extensions_table.h
There are other options.  We should pick one and document it.  Right
now, this file is chaos.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

intel/compiler: Implement untyped atomic float min, max, and compare-swap dataport messages

v2: Split changes to the message type field to another patch.  Suggested
by Caio.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

intel/compiler: Expand untyped atomic message type field by a bit

This is necessary for a new Gen9 message type that will be added in the
next patch.  There are also Gen8 message types that need the extra bit
(mostly for bindless).

v2: Split off from the next patch.  Suggested by Caio.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

intel/compiler: Silence unused parameter warnings

src/intel/compiler/brw_disasm_info.c: In function ‘nir_print_instr’:
src/intel/compiler/brw_disasm_info.c:30:61: warning: unused parameter ‘instr’ [-Wunused-parameter]
 __attribute__((weak)) void nir_print_instr(const nir_instr *instr, FILE *fp) {}
                                                             ^~~~~
src/intel/compiler/brw_disasm_info.c:30:74: warning: unused parameter ‘fp’ [-Wunused-parameter]
 __attribute__((weak)) void nir_print_instr(const nir_instr *instr, FILE *fp) {}
                                                                          ^~
src/intel/compiler/brw_disasm.c: In function ‘src_ia1’:
src/intel/compiler/brw_disasm.c:850:18: warning: unused parameter ‘_reg_file’ [-Wunused-parameter]
         unsigned _reg_file,
                  ^~~~~~~~~
src/intel/compiler/brw_fs_surface_builder.cpp: In function ‘void brw::surface_access::emit_byte_scattered_write(const brw::fs_builder&, const fs_reg&, const fs_reg&, const fs_reg&, unsigned int, unsigned int, unsigned int, brw_predicate)’:
src/intel/compiler/brw_fs_surface_builder.cpp:193:57: warning: unused parameter ‘size’ [-Wunused-parameter]
                                 unsigned dims, unsigned size,
                                                         ^~~~

v2: Update commit message.  brw_fs_generator.cpp warnings were already
fixed by another patch.  Noticed by Caio.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

nir: Add floating point atomic min, max, and compare-swap instrinsics

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

nir: Add floating point atomic add instrinsics

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

glsl: Add support for lowering shared-variable float atomics

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

glsl: Add support for lowering SSBO float atomics

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

glsl: Add built-in functions for INTEL_shader_atomic_float_minmax

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

mesa: Extension boilerplate for INTEL_shader_atomic_float_minmax

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

docs: Initial version of INTEL_shader_atomic_float_minmax spec

v2: Describe interactions with the capabilities added by
SPV_INTEL_shader_atomic_float_minmax

v3: Remove 64-bit float support.

v4: Explain NaN issues.  Explain issues with atomicMin(-0, +0) and
atomicMax(-0, +0).

v5: Fix whitespace issues noticed by Caio.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

glsl: Add built-in functions for NV_shader_atomic_float

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

mesa: Extension boilerplate for NV_shader_atomic_float

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

meson: fix egl build for android

Haven't tested this, but we do include loader.h
in platform_android.c

Fixes: c5ec1556859b7d33637c9fad13d3473c7b2f9eb3 ("meson: wire up egl/android")
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: fix egl build for surfaceless

Without this, I get:

 > platform_surfaceless.c:38:10: fatal error: 'loader.h' file not found
 > #include "loader.h"
 >      ^~~~~~~~~~
 > 1 error generated.

Fixes: 108d257a16859898f5ce02f4759c5c58f9b8c050 ("meson: build libEGL")
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
v2: Split up patches, modify commit message (Dylan)

nir: Give end_block its own index

Since there's no particular reason for the index to be 0, choose an
index that is not used by other block.  This is convenient when we
store "per-block" data in an array AND look for the successors
data (e.g. any kind of backwards data-flow analysis).

v2: Add a note about end_block's index. (Jason)

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

nir: Skip common instructions when comparing deref paths

Deref paths may share the same deref instructions in their chains,
e.g.

    ssa_100 = deref_var A
    ssa_101 = deref_struct "array_field" of ssa_100
    ssa_102 = deref_array "[1]" of ssa_101
    ssa_103 = deref_struct "field_a" of ssa_102
    ssa_104 = deref_struct "field_a" of ssa_103

when comparing the two last deref instructions, their paths will share
a common sequence ssa_100, ssa_101, ssa_102.  This patch skips to next
iteration if the deref instructions are the same.  Path[0] (the var)
is still handled specially, so in the case above, only ssa_101 and
ssa_102 will be skipped.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

nir: Export deref comparison functions

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

util/dynarray: add a clone function

v2: Fix mem_ctx parameter type. (Thomas)

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

amd/addrlib: Fix include path for c99_compat.h

Without this patch mesa doesn't compile:

In file included from ../mesa-9999/src/amd/addrlib/addrinterface.cpp:39:
../mesa-9999/src/util/macros.h:29:10: fatal error: c99_compat.h: No such file or directory
 #include "c99_compat.h"
          ^~~~~~~~~~~~~~
compilation terminated.

Fixes: 15ca5ce99a80d9ebb5ef2b1aca6ea00784931de4
       ("amd/addrlib: mark returnCode as MAYBE_UNUSED in")
Signed-off-by: Mariusz Ceier <mceier+mesa-dev@gmail.com>
Acked-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

vulkan/wsi: fix pointer-integer conversion warnings

For 32bit build. Trivial.

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

radv: use different builtin shader cache for 32bit

Currently if 64bit and 32bit programs are used interchangeably, radv
will keep overwriting the cache. Use separate cache files to avoid
that.

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

radv: place pointer length into cache uuid

Thanks to reproducible builds, binary file timestamps may be identical
for both 32bit and 64bit packages when built from the same source.
This means radv will use the same cache for both 32 and 64 bit
processes, which leads to crashes.

Conveniently there is a spare byte in cache_uuid, let's place the
pointer size there.

Fixes: f4e499ec79 "radv: add initial non-conformant radv vulkan driver"
CC: 18.1 18.2 <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107601
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105904
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

llvmpipe: add cc clobber to inline asm

The bsr instruction modifies flags, so that needs to be indicated to the
compiler. No effect on generated code, but still needed for correctness.

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

intel/isl: Avoid tiling some 16K-wide render targets

Fix rendering issues on BDW and SKL.

Fixes: 0288fe8d0417730bdd5b3477130dd1dc32bdbcd3
("i965/miptree: Use the correct BLT pitch")

Fixes the following regressions seen

exclusively on SKL:
* KHR-GL46.texture_barrier_ARB.disjoint-texels
* KHR-GL46.texture_barrier_ARB.overlapping-texels
* KHR-GL46.texture_barrier.disjoint-texels
* KHR-GL46.texture_barrier.overlapping-texels

and both on BDW and SKL:
* GTF-GL46.gtf21.GL2FixedTests.buffer_corners.buffer_corners
* GTF-GL46.gtf21.GL2FixedTests.stencil_plane_corners.stencil_plane_corners

v2: Note the fixed tests (Andres).
    Don't cause failures with multisampled buffers (Andres).
    Don't hamper SKL GT4 (Ken).
v3: Fix the Fixes tag (Dylan).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107359
Cc: <mesa-stable@lists.freedesktop.org>
Tested-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965/miptree: Fix can_blit_slice()

Check the destination's row pitch against the BLT engine's row pitch
limitation as well.

Fixes: 0288fe8d0417730bdd5b3477130dd1dc32bdbcd3
("i965/miptree: Use the correct BLT pitch")

v2: Fix the Fixes tag (Dylan).
    Check the destination row pitch (Chris).

Reported-by: Dylan Baker <dylan@pnwbakers.com>
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965/miptree: Use miptree_map in map_blit functions

This struct contains all the data of interest. can_blit_slice() will use
it in the next patch to calculate the correct pitch.

Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

intel/tools/aubwrite: Always use physical addresses for traces.

It looks like we can't rely on the simulator to always translate virtual
addresses to physical ones correctly. So let's use physical everywhere.

Since our current GGTT maps virtual to physical addresses in a 1:1 way,
no further changes are required.

Additionally, we have other address spaces not in use right now. So
let's make it easier to switch which one we are using but putting the
default one into the aub_file struct.

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

intel/tools/aubwrite: Rename "legacy" to "Trace Block".

Hopefully it's a little more descriptive, and more accurate.

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

nir/vars_to_ssa: Don't build deref nodes for non-local variables

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

ac: fix WAITCNT flags for GFX9

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

amd/addrlib: mark physicalSliceSize as MAYBE_UNUSED in Addr::V1::EgBasedLib::HwlGetSizeAdjustmentMicroTiled

Only used, when asserts are enabled.

Fixes an unused-but-set-variable warning with GCC 8:
 ../../../src/amd/addrlib/r800/egbaddrlib.cpp: In member function 'virtual long long unsigned int Addr::V1::EgBasedLib::HwlGetSizeAdjustmentMicroTiled(unsigned int, unsigned int, ADDR_SURFACE_FLAGS, unsigned int, unsigned int, unsigned int, unsigned int*, unsigned int*) const':
 ../../../src/amd/addrlib/r800/egbaddrlib.cpp:4111:13: warning: variable 'physicalSliceSize' set but not used [-Wunused-but-set-variable]
      UINT_64 physicalSliceSize;
              ^~~~~~~~~~~~~~~~~

Signed-off-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

amd/addrlib: mark numPipes as MAYBE_UNUSED in Addr::V1::EgBasedLib::SanityCheckMacroTiled (v2)

Only used, when asserts are enabled.

Fixes an unused-variable warning with GCC 8:
 ../../../src/amd/addrlib/r800/egbaddrlib.cpp: In member function 'int Addr::V1::EgBasedLib::SanityCheckMacroTiled(ADDR_TILEINFO*) const':
 ../../../src/amd/addrlib/r800/egbaddrlib.cpp:982:13: warning: unused variable 'numPipes' [-Wunused-variable]
      UINT_32 numPipes    = HwlGetPipes(pTileInfo);
              ^~~~~~~~

v2: Don't realign other variable definitions, to keep in line with file
    style (Marek)

Signed-off-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

amd/addrlib: mark *pEqToCheck as MAYBE_UNUSED in Addr::V2::Gfx9Lib::ComputeStereoInfo (v2)

Only used, when asserts are enabled.

Fixes an unused-variable warning with GCC 8:
 ../../../src/amd/addrlib/gfx9/gfx9addrlib.cpp: In member function 'ADDR_E_RETURNCODE Addr::V2::Gfx9Lib::ComputeStereoInfo(const ADDR2_COMPUTE_SURFACE_INFO_INPUT*, ADDR2_COMPUTE_SURFACE_INFO_OUTPUT*, unsigned int*) const':
 ../../../src/amd/addrlib/gfx9/gfx9addrlib.cpp:3879:34: warning: unused variable 'pEqToCheck' [-Wunused-variable]
              const ADDR_EQUATION *pEqToCheck        = &m_equationTable[eqIndex];
                                   ^~~~~~~~~~

v2: Don't realign other variable definitions, to keep in line with file
    style (Marek)

Signed-off-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

amd/addrlib: mark microBlockDim as MAYBE_UNUSED in Addr::V2::Gfx9Lib::HwlComputeBlock256Equation

Only used, when asserts are enabled.

Fixes an unused-but-set-variable warning with GCC 8:
 ../../../src/amd/addrlib/gfx9/gfx9addrlib.cpp: In member function 'virtual ADDR_E_RETURNCODE Addr::V2::Gfx9Lib::HwlComputeBlock256Equation(AddrResourceType, AddrSwizzleMode, unsigned int, ADDR_EQUATION*) const':
 ../../../src/amd/addrlib/gfx9/gfx9addrlib.cpp:2473:15: warning: variable 'microBlockDim' set but not used [-Wunused-but-set-variable]
          Dim2d microBlockDim = Block256_2d[elementBytesLog2];
                ^~~~~~~~~~~~~

Signed-off-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

amd/addrlib: mark returnCode as MAYBE_UNUSED in ElemGetExportNorm

Only used, when asserts are enabled.

Fixes an unused-but-set-variable warning with GCC 8:
 ../../../src/amd/addrlib/addrinterface.cpp: In function 'int ElemGetExportNorm(ADDR_HANDLE, const ELEM_GETEXPORTNORM_INPUT*)':
 ../../../src/amd/addrlib/addrinterface.cpp:835:23: warning: variable 'returnCode' set but not used [-Wunused-but-set-variable]
      ADDR_E_RETURNCODE returnCode = ADDR_OK;
                        ^~~~~~~~~~

Signed-off-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

intel: aubinator_viewer: add urb view

This is available through a "Show URB" button on the 3DPRIMITIVE
instructions.

v2: Fix urb allocation end value in tooltip (Rafael)

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

intel: aubinator_viewer: store urb state during decoding

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

intel: tools: add aubinator viewer

A graphical user interface version of aubinator.
Allows you to :

   - simultaneously look at multiple points in the aub file (using all
     the goodness of the existing decoding in aubinator)

   - edit an aub file

v2: Switch from GLFW to GTK+3

v3: Fix warning when exiting

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Rafael Antognolli <rafael.antognolli@intel.com> (v1)

intel: tools: import ImGui

We want to add a new UI tool to decode aub files. This will use the
Dear ImGui library to render its interface. The build of this UI
toolkit is conditional to -Dwith_tools=intel-ui which superseeds
-Dwith_tools=intel.

The main way to use ImGui is to embed its source code at a particular
revision. Most embedding projects have to do a bit of integration
which is really specific to one's project. In our case the only
modification is to include libepoxy. We also choose to use Gtk+3 for
the window system integration. As oppose to the previous previous
version of this patch using GLFW, Gtk+ is able to handle X11/Wayland
session as well as property DPI scaling on retina monitors.

The import was done at this commit (https://github.com/ocornut/imgui) :

commit 6211f40f3d903dd9df961256e044029c49793aa3
Author: omar <omarcornut@gmail.com>
Date:   Fri Jul 27 12:29:33 2018 +0200

    Internals: Drag and Drop: default drop preview use a narrower clipping rectangle (no effect here, but other branches uses a narrow clipping rectangle that was too small so this is a fix for it) + Comments

v2: Switch from GLFW to GTK+ (Lionel)

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

intel: tools: aub_mem: reuse already mapped ppgtt buffers

When we map a PPGTT buffer into a continous address space of aubinator
to be able to inspect it, we currently add it to the list of BOs to
unmap once we're finished. An optimization we can apply it to look up
that list before trying to remap PPGTT buffers again (we already do
this for GGTT buffers).

We need to take some care before doing this because the list also
contains GGTT BOs. As GGTT & PPGTT are 2 different address spaces, we
can have matching addresses in both that point to different physical
locations.

This changes adds a flag on the elements of the list of mapped BOs to
differenciate between GGTT & PPGTT, which allows use to reuse that
list when looking up both address spaces.

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

intel: tools: aubmem: map gtt data to aub file

This will allow the aubinator viewer tool to modify the aub data that
was loaded at a particular gtt address.

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

intel: tools: create libaub

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

intel: tools: aubwrite: wrap function declarations for c++

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

intel: tools: split memory management out of aubinator

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

util: rb_tree: add safe iterators

v2: Add helper to make iterators more readable (Rafael)
    Fix rev iterator bug (Rafael)

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

intel: tools: split aub parsing from aubinator

v2: add parsing error callback (Lionel)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com> (v1)

meson: Run the test with Python 3

This is a patch from me and a patch from Mathieu Bridon squashed
together.

Signed-off-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Mathieu Bridon <bochecha@daitauha.fr>

python: Disable universal newlines

We are testing the behaviour of a tool, for different input files, each
one using a different newline sequence. ('\n' on UNIX, '\r\n' on
Windows, …)

Unfortunately, when opening a file in text mode, Python 3 will by
default enable the "universal newlines" mode, which means it replaces
all the known newline sequences by '\n'.

This (usually useful) behaviour breaks the tests, which are specifically
trying to handle files with newline sequences different from '\n'.

Disabling the universal newlines mode fixes the tests.

However, to keep the script compatible with both Python 2 and 3, we must
use the io.open() function instead of the open() builtin, as the latter
only knows about the `newline` argument on Python 3.

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

python: difflib prefers unicode strings

Python 3 does not automatically convert from bytes to unicode strings
like Python 2 used to do.

This commit makes sure we pass unicode strings to difflib.unified_diff,
so that the script works on both Python 2 and 3.

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

compiler/glsl/tests: Make tests python3 safe

v2: - explicitly decode the output of subprocesses
    - handle bytes and string types consistently rather than relying on
      python 2's coercion for bytes and ignoring them in python 3
v3: - explicitly set encode as well as decode
    - python 2.7 and 3.x `bytes` instead of defining an alias

Reviewed-by: Mathieu Bridon <bochecha@daitauha.fr>

travis: SWR requires LLVM 6.0

v2: update clarification why ubuntu-toolchain-r-test is required (Emil)

Fixes: 0cef0cccf51 ("swr: bump minimum supported LLVM version to 6.0")
Cc: Dylan Baker <dylan@pnwbakers.com>
Cc: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

ac/nir: fix getting GLSL type of array of samplers for TG4

This fixes a crash in build_tex_intrinsic() when trying to
launch the Basemark GPU benchmark on GFX8. It looks like
there is still something wrong because some frames are black.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106980
CC: 18.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

radv: remove dead variables after splitting per member structs

Otherwise, nir_lower_clip_cull_distance_arrays might report
wrong number of output clips/culls because it relies on
shader output variables and some of them might be dead.

This fixes a rendering issue with Dolphin and Super Mario
Sunshine.

Fixes: b0c643d8f5 ("spirv: Use NIR per-member splitting")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107610
CC: 18.2 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

anv: add VK_EXT_sampler_filter_minmax support

This extension can be supported on SKL+. With this patch,
all corresponding tests (6K+) in CTS can pass. No test fails.

I verified CTS with the command below:
deqp-vk --deqp-case=dEQP-VK.pipeline.sampler.view_type.*reduce*

v2: 1) support all depth formats, not depth-only formats, 2) fix
a wrong indention (Jason).

v3: fix a few nits (Lionel).

v4: fix failures in CI: disable sampler reduction when sampler
reduction mode is not specified via this extension (Lionel).

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

radv: use ac_build_imad()

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

ac,radeonsi: use ac_build_gather_values more

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

ac,radeonsi: use ac_build_fmad

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

radeonsi: use ac_build_imad

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

ac: add imad & fmad helpers

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

ac: add ac_build_s_barrier

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

radeonsi: print the shader stage name when printing LLVM IR

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

radeonsi: use is_merged shader in si_prolog_get_rw_buffers

needed to change the input type to si_shader_context

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

ac: completely remove +auto-waitcnt-before-barrier

it causes corruption on several different GPU generations.

Cc: 18.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

anv/icl: Allow headerless sampler messages for pre-emptable contexts

It fixes simulator warnings in vulkancts tests complaining about missing
support for headerless sampler messages for pre-emptable contexts.
Bit 5 in SAMPLER MODE register is newly introduced for ICLLP.

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