nir: add two new texture ops for multisample fragment color/mask fetches

This introduces:
   - nir_texop_fragment_mask_fetch (fetch a fragment mask from a
     compressed multisampled color surface)
   - nir_texop_fragment_fetch (fetch a color fragment for a
     particular sample at corresponding fragment mask index).

These two texture operations are necessary for implementing
SPV_AMD_shader_fragment_mask.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3304>

spirv: add SpvCapabilityFragmentMaskAMD

This new capability is for SPV_AMD_shader_fragment_mask.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3304>

radv: handle missing implicit subpass dependencies

When a subpass doesn't declare an explicit dependency from/to
VK_SUBPASS_EXTERNAL, Vulkan says there is an implicit dependency.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3330>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3330>

radv: add explicit external subpass dependencies to meta operations

No functional changes because a subpass dependency with dstStageMask
set to VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT is a no-op.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3330>

gallivm: fix find lsb

the GLSL return value is different than the llvm intrinsic.

Fixes arb gpu shader5 tests

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3528>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3528>

galllivm: fix gather offset casting

cast texture offsets to 32-bit integers

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3528>

llvmpipe: fix some integer instruction lowering.

We want to lower to shifts for bitfields, and lower ifind_msb.

Fixes a bunch of gpu shader5 tests.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3528>

gallivm: fix gather component handling.

Fixes the extended gather test for gpu shader5

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3528>

turnip: Add support for uniform texel buffers.

Pretty straightforward: Port texture descriptor code from freedreno, fill
in alignment limits from closed vk, and tu_cmd_buffer.c was already
uploading the texture descriptor.

This doesn't implement storage texel buffers (required in the compute
pipeline) yet, since those will need an IBO descriptor for the store path.
Still, making the load path be connected to the texture descriptor won't
hurt.

Part of #2237

Fixes dEQP-VK.binding_model.shader_access.primary_cmd_buf.uniform_texel_buffer.*

Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3522>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3522>

intel: Fix aux map alignments on 32-bit builds.

ALIGN() brilliantly uses uintptr_t, making it unsafe for use with 64-bit
GPU addresses in 32-bit builds of the driver.  Use align64() instead,
which uses uint64_t.

Fixes assertion failures when running any 32-bit program on Tigerlake.

Fixes: 2e6a7ced4db ("iris/gen12: Write GFX_AUX_TABLE base address register")
Fixes: 0d0290bb3f7 ("intel/common: Add surface to aux map translation table support")
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3507>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3507>

util: Remove tmp argument from BITSET_FOREACH_SET macro

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3499>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3499>

util: Explain BITSET_FOREACH_SET params

__size, in particular, makes this macro rather confusing to understand
how to use. Hopefully this comment saves future users the headache.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3499>

lima: implement invalidate_resource()

We don't need to resolve invalidated resources, so it should
improve performance for applications that are doing this hint.

Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3476>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3476>

glsl_to_nir: update interface type properly

Since 76ba225184ed the member variable types were being redefined
but we assigned the old interface type to the variable.

In a following patch series we will use the types to check if we
are dealing with an interface instance when apply GLSL linking
rules.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3468>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3468>

glsl: count uniform components and storage better in nir linking

This helps avoid incorrect validation error when linking glsl
shaders and avoids assigning uniform storage slots that will
never be used.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3468>

glsl: fix check for matrices in blocks when using nir uniform linker

We need to stripe any arrays before checking the type. Here we
just use the uniform type which has already be stripped.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3468>

glsl: remove bogus assert in nir uniform linking

I'm not sure why this was first added but it causes an assert
on any uniform matrix.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3468>

nir/algebraic: Optimize some 64-bit integer comparisons involving zero

I noticed that we can do better for these kinds of comparisons while
working on the lowering for iadd_sat@64 and isub_sat@64.  This
eliminated 11 instruction from the fs-addSaturate-int64.shader_test.

My hope is that this will improve the run-time of int64 tests on Ice
Lake.  I have no data to support or refute this.

Unsurprisingly, no changes on shader-db.

v2: Condition the min and max patterns with nir_lower_minmax64.
Suggested by Caio.  Very long discussion in the MR. :)

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

anv: Enable SPV_INTEL_shader_integer_functions2 and VK_INTEL_shader_integer_functions2

Currently only implemented in the scalar backend, so only enable for
Gen8+.  If support for the other opcodes is added to the vec4 backend,
Gen7 could be supported.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

iris: Enable INTEL_shader_integer_functions2

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

gallium: Add a cap bit for integer multiplication between 32-bit and 16-bit

Driver supports integer multiplication between a 32-bit integer and a
16-bit integer.  If the second operand is 32-bits, the upper 16-bits are
ignored, and the low 16-bits are possibly sign extended as necessary.

Iris will eventually enable this.  Not sure about other drivers.

v2: Add default value to u_screen.c.  Suggested by Caio.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

gallium: Add a cap bit for OpenCL-style extended integer functions

Iris will eventually enable this.  Looking at the header files, it looks
like Midgard could also enable it.  Basically, any GPU that fully
supports OpenCL can.

v2: Add default value to u_screen.c.  Suggested by Caio.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

i965: Enable INTEL_shader_integer_functions2 on Gen8+

v2: Use new lower_hadd64 and lower_usub_sat64 flags.

v3: Enable SPIR-V capability.

v4: Move lowering options to COMMON_SCALAR_OPTIONS.  Suggested by Caio.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

spirv: Add support for IntegerFunctions2INTEL capability

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

spirv: Silence a bunch of unused parameter warnings

The change to get_uniform_nir_atomic_op make it look like the other
get_*_nir_atomic_op functions.  The rest just add UNUSED or ASSERTED
to parameters required for some of the interfaces.

src/compiler/spirv/spirv_to_nir.c: In function ‘struct_member_decoration_cb’:
src/compiler/spirv/spirv_to_nir.c:673:47: warning: unused parameter ‘val’ [-Wunused-parameter]
                             struct vtn_value *val, int member,
                                               ^~~
src/compiler/spirv/spirv_to_nir.c: In function ‘struct_member_matrix_stride_cb’:
src/compiler/spirv/spirv_to_nir.c:778:50: warning: unused parameter ‘val’ [-Wunused-parameter]
                                struct vtn_value *val, int member,
                                                  ^~~
src/compiler/spirv/spirv_to_nir.c: In function ‘type_decoration_cb’:
src/compiler/spirv/spirv_to_nir.c:805:61: warning: unused parameter ‘ctx’ [-Wunused-parameter]
                     const struct vtn_decoration *dec, void *ctx)
                                                             ^~~
src/compiler/spirv/spirv_to_nir.c: In function ‘spec_constant_decoration_cb’:
src/compiler/spirv/spirv_to_nir.c:1359:70: warning: unused parameter ‘v’ [-Wunused-parameter]
 spec_constant_decoration_cb(struct vtn_builder *b, struct vtn_value *v,
                                                                      ^
src/compiler/spirv/spirv_to_nir.c: In function ‘handle_workgroup_size_decoration_cb’:
src/compiler/spirv/spirv_to_nir.c:1407:43: warning: unused parameter ‘data’ [-Wunused-parameter]
                                     void *data)
                                           ^~~~
src/compiler/spirv/spirv_to_nir.c: In function ‘vtn_handle_function_call’:
src/compiler/spirv/spirv_to_nir.c:1806:55: warning: unused parameter ‘opcode’ [-Wunused-parameter]
 vtn_handle_function_call(struct vtn_builder *b, SpvOp opcode,
                                                       ^~~~~~
src/compiler/spirv/spirv_to_nir.c:1807:54: warning: unused parameter ‘count’ [-Wunused-parameter]
                          const uint32_t *w, unsigned count)
                                                      ^~~~~
src/compiler/spirv/spirv_to_nir.c: In function ‘get_uniform_nir_atomic_op’:
src/compiler/spirv/spirv_to_nir.c:2548:47: warning: unused parameter ‘b’ [-Wunused-parameter]
 get_uniform_nir_atomic_op(struct vtn_builder *b, SpvOp opcode)
                                               ^
src/compiler/spirv/spirv_to_nir.c: In function ‘vtn_handle_atomics’:
src/compiler/spirv/spirv_to_nir.c:2633:48: warning: unused parameter ‘count’ [-Wunused-parameter]
                    const uint32_t *w, unsigned count)
                                                ^~~~~
src/compiler/spirv/spirv_to_nir.c: In function ‘vtn_handle_barrier’:
src/compiler/spirv/spirv_to_nir.c:3197:48: warning: unused parameter ‘count’ [-Wunused-parameter]
                    const uint32_t *w, unsigned count)
                                                ^~~~~
src/compiler/spirv/spirv_to_nir.c: In function ‘vtn_handle_execution_mode’:
src/compiler/spirv/spirv_to_nir.c:3618:68: warning: unused parameter ‘data’ [-Wunused-parameter]
                           const struct vtn_decoration *mode, void *data)
                                                                    ^~~~

Acked-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

nir/spirv: Translate SPIR-V to NIR for new INTEL_shader_integer_functions2 opcodes

v2: Rebase on 272e927d0e9 ("nir/spirv: initial handling of OpenCL.std
extension opcodes")

v3: Add missing SpvOpUCountTrailingZerosINTEL case to switch in
vtn_handle_body_instruction. Remove stray semicolon in
vtn_nir_alu_op_for_spirv_opcode. Use umin instead of umax for
SpvOpUCountTrailingZerosINTEL "lowering" in vtn_handle_alu.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

intel/fs: Implement support for NIR opcodes for INTEL_shader_integer_functions2

v2: Remove smashing type to D for nir_op_irhadd.  Caio noticed it was
odd, and removing it fixes an assertion failure in the crucible
func.shader.averageRounded.int64_t test (because the source should be
W).

v3: Emit BRW_OPCODE_MUL directly for nir_op_umul_32x16 and
nir_op_imul_32x16.  Suggested by Curro.

v4: Smash types of MUL instruction generated for nir_op_umul_32x16 and
nir_op_imul_32x16.  With this change, I get the same assembly now as I
did with v2.

v5: Remove support for pre-Gen7.  The integer multiply path was
incorrect, and, since the extension isn't enabled pre-Gen7, there's no
way to test it.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

intel/fs: Add SHADER_OPCODE_[IU]SUB_SAT pseudo-ops

v2: Add a big comment explaining the [IU]SUB_SAT lowering.  Suggested by
Caio.

v3: Use get_fpu_lowered_simd_width in get_lowered_simd_width.  Suggested
by Ken on IRC.

v4: Fix a typo in a comment.  Noticed by Caio.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

intel/fs: Don't lower integer multiplies that don't need lowering

v2: Move the check to fs_visitor::lower_integer_multiplication.
Previously the cases where lowering was skipped, the original
instruction was removed by fs_visitor::lower_integer_multiplication.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

compiler: Translate GLSL IR to NIR for new INTEL_shader_integer_functions2 expressions

v2: Rebase on 272e927d0e9 ("nir/spirv: initial handling of OpenCL.std
extension opcodes")

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

nir/algebraic: Add lowering for 64-bit iadd_sat and isub_sat

v2: Rearranged and expand the comment about the optimizations applied to
the lowering.  Suggested by Caio.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

nir/algebraic: Add lowering for 64-bit uadd_sat

Fixes piglit fs-addsaturate-uint64 and vs-addsaturate-uint64 on Ice
Lake.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

nir/algebraic: Add lowering for 64-bit usub_sat

v2: Rebase on 272e927d0e9 ("nir/spirv: initial handling of OpenCL.std
extension opcodes")

v3: Add a new lower_usub_sat64 flag that only applies to the 64-bit
version of the nir_op_usub_sat instruction.

v4: Also enable the lowering when nir_lower_iadd64 is set.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com> [v3]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

nir/algebraic: Add lowering for 64-bit hadd and rhadd

v2: Rebase on 272e927d0e9 ("nir/spirv: initial handling of OpenCL.std
extension opcodes")

v3: Add a new lower_hadd64 flag that only applies to the 64-bit versions
of the instructions.

v4: Also enable the lowering when nir_lower_iadd64 is set.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com> [v3]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

nir/algebraic: Add lowering for uabs_usub and uabs_isub

v2: Remove some rebase failures noticed by Caio.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

nir: Add new instructions for INTEL_shader_integer_functions2

uctz isn't added because it will implemented in the GLSL path and the
SPIR-V path using other pre-existing instructions.

v2: Avoid signed integer overflow for uabs_isub(0, INT_MIN).  Noticed by
Caio.

v3: Alternate fix for signed integer overflow for abs_sub(0, INT_MIN).
I tried the previous methon in a small test program with -ftrapv, and it
failed.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

glsl: Add built-in functions for INTEL_shader_integer_functions2

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

glsl_types: Add function to get an unsigned base type from a signed type

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

glsl: Add new expressions for INTEL_shader_integer_functions2

v2: Re-write iadd64_saturate and isub64_saturate to avoid undefined
overflow behavior.  Also fix copy-and-paste bug in isub64_saturate.
Suggested by Caio.

v3: Avoid signed integer overflow for abs_sub(0, INT_MIN).  Noticed by
Caio.

v4: Alternate fix for signed integer overflow for abs_sub(0, INT_MIN).
I tried the previous methon in a small test program with -ftrapv, and it
failed.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

mesa: Extension boilerplate for INTEL_shader_integer_functions2

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

intel/compiler: Move Gen4/5 rounding to visitor

Gen4/5's rounding instructions operate differently than later Gens'.
They all return the floor of the input and the "Round-increment"
conditional modifier answers whether the result should be incremented by
1.0 to get the appropriate result for the operation (and thus its
behavior is determined by the round opcode; e.g., RNDZ vs RNDE).

Since this requires a second instruciton (a predicated ADD) that
consumes the result of the round instruction, the round instruction
cannot write its result directly to the (write-only) message registers.
By emitting the ADD in the generator, the backend thinks it's safe to
store the round's result directly to the message register file.

To avoid this, we move the emission of the ADD instruction to the NIR
translator so that the backend has the information it needs.

I suspect this also fixes code generated for RNDZ.SAT but since
Gen4/5 don't support GLSL 1.30 which adds the trunc() function, I
couldn't write a piglit test to confirm. My thinking is that if x=-0.5:

      sat(trunc(-0.5)) = 0.0

But on Gen4/5 where sat(trunc(x)) is implemented as

      rndz.r.f0  result, x             // result = floor(x)
                                       // set f0 if increment needed
      (+f0) add  result, result, 1.0   // fixup so result = trunc(x)

then putting saturate on both instructions will give the wrong result.

      floor(-0.5) = -1.0
      sat(floor(-0.5)) = 0.0
      // +1 increment would be needed since floor(-0.5) != trunc(-0.5)
      sat(sat(floor(-0.5)) + 1.0) = 1.0

Fixes: 6f394343b1f ("nir/algebraic: i2f(f2i()) -> trunc()")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2355
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3459>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3459>

meson: Do not require libdrm for DRI2 on hurd

Cc: 19.3 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3231>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3231>

util: Do not fail to build on unknown pthread_setname_np

This is only used for debugging, so better making porting on various systems
less hard.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3229>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3229>

loader: #define PATH_MAX when undefined (eg. Hurd)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3228>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3228>

util/atomic: fix return type of p_atomic_add_return() fallback

Fixes: 385d13f26d2b69db9423 ("util/atomic: Add a _return variant of p_atomic_add")
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3012>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3012>

gallium: dmabuf support for yuv formats that are not natively supported

V2 (Kenneth Graunke):
   added a helper function to check if every format is supported

Signed-off-by: James Xiong <james.xiong@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2846>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2846>

intel/compiler: Return early if read() failed

This was the only warning I could see while compiling Iris.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2821>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2821>

intel/perf: adapt to platforms like Solaris without d_type in struct dirent

Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
[Eric: factor out the is_dir_or_link() check and fix a bug in v1]
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
v3: include directory path when lstat'ing files
v4: fix inverted check in enumerate_sysfs_metrics()

Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2258>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2258>

llvmpipe: drop LLVM < 3.4 support

We don't support < 3.9 anymore, so this code is dead.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2760>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2760>

egl: drop confusing mincore() error message

A user came to me asking how to fix this error, but it's entirely
expected that `get_wl_surface_proxy()` on recent enough wayland
compositors will always print it.
Let's just remove the message altogether, it is basically never useful.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3219>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3219>

aco: fix off-by-one error when initializing sgpr_live_in

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2394
Fixes: 93c8ebfa780 ('aco: Initial commit of independent AMD compiler')
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3511>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3511>

radv: fix double free corruption in radv_alloc_memory()

If the driver fails to allocate memory for some reasons, it shouldn't
free the 'mem' object twice.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2302
Fixes: 825ddfee599 ("radv: Handle device memory alloc failure with normal free.")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3508>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3508>

gitlab-ci: Use single if for manual job rules entry

I thought multiple ifs would all need to match, but looks like only the
last one (or either one?) does.

This should prevent a manual pipeline from getting created after merging
changes which can't affect the pipeline.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3474>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3474>

gitlab-ci: Set GIT_STRATEGY to none for the dummy job

It doesn't need anything from the Git repository.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3474>

util/u_thread: Fix build under Haiku

haiku/hgl: Fix build via header reordering

aco: fix operand kill flags when a temporary is used more than once

Helps create v_mac_f32 from v_mad_f32(b, a, b)

Totals from affected shaders:
SGPRS: 35824 -> 35824 (0.00 %)
VGPRS: 33460 -> 33456 (-0.01 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 2187264 -> 2180976 (-0.29 %) bytes
LDS: 127 -> 127 (0.00 %) blocks
Max Waves: 3802 -> 3802 (0.00 %)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3486>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3486>

panfrost/midgard: Add missing lowering passes for type/size conversion ops

Replace the manual type/size conversion lowering description by one
that's automatically generated and covers all type/size conversions.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

panfrost/midgard: Add 64 bits float <-> int converters

The 64 bit converter cases were missing, add them now.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

panfrost/midgard: Fix mir_print_instruction() for branch instructions

Branch instructions should not be treated as regular ALUs.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

panfrost/midgard: Add f2f64 support

So we can convert floats into doubles.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

panfrost/midgard: Factorize f2f and u2u handling

Those size conversion operations work the same way apart from f2f
using an fmov op code and u2u using an imov. Let's handle them in the
same case block to avoid code duplication.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

panfrost/midgard: Make sure promote_fmov() only promotes 32-bit imovs

mir_constant_float() assumes we're dealing with 32-bit integers/floats,
which is only the case if reg_mode is equal to midgard_reg_mode_32.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

panfrost/midgard: Rework mir_adjust_constants() to make it type/size agnostic

Right now, constant combining is not supported in 16 bit mode, and 64
bit mode is simply ignored. Let's rework the function to make it
type/bit-size agnostic.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

panfrost/midgard: Use a union to manipulate embedded constants

Each instruction bundle can contain up to 16 constant bytes. The meaning
of those byte is instruction dependent: it depends on the instruction
native type (int, uint or float) and the instruction reg_mode (8, 16, 32
or 64 bit). Those different layouts can be exposed as a union to
facilitate constants manipulation.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3478>

anv: ensure prog params are initialized with 0s

As a result of 9baa33cef01f our backend compiler leaves params pretty
much untouched. So in order to avoid storing uninitialized values in
the shader cache blobs, just 0 out this array.

I've considered not even allocating this array which works on gen8+
but the vec4 backend still makes a copy of this array and so it
crashes on memcpy on HSW.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 9baa33cef01f ("anv: Rework push constant handling")
Reported-by: Tapani Pälli <tapani.palli@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Acked-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3516>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3516>

panfrost: Fix crash in compute variant allocation

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Fixes: d8a3501f1b2 ("panfrost: Dynamically allocate shader variants")
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3515>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3515>

etnaviv: drm: Don't miscalculate timeout

The current code overflows (s * 1000000000) for s >= 5 but that is
e.g. used in etna_bo_cpu_prep.

Signed-off-by: Guido Günther <agx@sigxcpu.org>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3509>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3509>

egl: Fix _eglPointerIsDereferencable w/o mincore()

On platforms without mincore(), _eglPointerIsDereferencable()
currently just checks whether p != NULL. This is not sufficient:
In the Wayland platform code (i.e., in get_wl_surface_proxy()),
_eglPointerIsDereferencable() is called on the version field
of `struct wl_egl_window` which is 3 on current versions of
Wayland. This causes a segfault when trying to dereference p.

Fix this behavior by assuming that the first page of the
process is never dereferencable.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3103>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3103>

egl/android: fix buffer_count for applications setting max count

Problem with previous solution was that it did not take account that
some applications may set a max count for buffers. Therefore we need to
query both min and max and clamp our setting based on that.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2373
Fixes: be08e6a4496 ("egl/android: Restrict minimum triple buffering for android color_buffers")
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3480>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3480>

aco: Fix signedness compare warning.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3483>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3483>

aco: Fix maybe-uninitialized warnings.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3483>

aco: Fix -Wstringop-overflow warnings in aco_span.

GCC does not understand how aco_span works.
This patch fixes it by casting the aco_span's this pointer
to uintptr_t rather than to a char pointer, effectively
telling GCC not to try to figure it out.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3483>

radeon: Fix multiple definition error with radeon_debug

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3488>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3488>

gallium: Fix a couple of multiple definition warnings.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3488>

r600: Move get_pic_param to radeon_vce.c

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3488>

radeon: Move si_get_pic_param to radeon_vce.c

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3488>

intel/compiler: Fix array bounds warning on GCC 10.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

turnip: Add support for non-zero (still constant) UBO buffer indices.

This was actually all ready to go at this point, and just needed to
increment by the value.

Fixes dEQP-VK.binding_model.shader_access.primary_cmd_buf.uniform_buffer.*

Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3504>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3504>

turnip: fix array/matrix varyings

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3109>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3109>

turnip: remove tu_sort_variables_by_location

nir_assign_io_var_locations already does sorting.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3109>

freedreno/ir3: allow inputs with the same location

turnip can have multiple inputs with the same location, and different
location_frac.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3109>

gitlab-ci: Skip ext_timer_query/time-elapsed

This test's result is unpredictable, so it may occasionally pass when we
expect it to fail, thus causing the CI pipeline to fail.

Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3498>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3498>

intel/compiler: Test compaction on Gen <= 12

With the previous commits we can now enable the unit test on Gen <= 12.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Validate fuzzed instructions

... before giving them to the instruction compactor.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Add unit tests for new EU validation checks

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Validate some instruction word encodings

Specifically, execution size, register file, and register type. I did
not add validation for vertical stride and width because I don't believe
it's possible to have an otherwise valid instruction with an invalid
vertical stride or width, due to all of the other regioning
restrictions.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Factor out brw_validate_instruction()

In order to fuzz test instructions, we first need to do some sanity
checking first. Factoring out this function allows us an easy way to
validate a single instruction.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Handle invalid compacted immediates

16-bit immediates need to be replicated through the 32-bit immediate
field, so we should never see one that isn't.

This does happen however in the fuzzer unit test, so returning false
allows the fuzzer to reject this case.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Handle invalid inputs to brw_reg_type_to_*()

Necessary to handle these cases when we test fuzzed instructions.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Split hw_type tables

Previously we were sharing tables between generations that were nearly
identical (i.e., Gen8 3-src adds HF support) and used a small bit of
code to handle the differences. This is kind of a mess if you want to
reject 64-bit types on platforms that don't support 64-bit types, so
split the tables, allowing each generation's table to list exactly what
it supports.

Acked-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Add a INVALID_{,HW_}REG_TYPE macros

Since the enum brw_reg_type is packed, comparisons with -1 don't work
directly, necessitating the cast. Add a macro to avoid this confusion.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Add NF some more places

Necessary to handle these cases when we test fuzzed instructions.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Limit compaction unit tests to specific gens

Two of the tests emit instructions with MRF destinations, and MRFs
aren't present on Gen7+. I think we were just lucky that this didn't
cause a problem earlier since we were running the tests on Gen7-9.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Don't disassemble align1 3-src operands on Gen < 10

Since the platforms don't support align1 3-src instructions, the
contents of these operands are not going to be meaningful. Just don't
print them to avoid hitting some assertions in brw_inst functions.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Split has_64bit_types into float/int

Gen7 has 64-bit floats but not 64-bit ints.

Acked-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Extract GEN_* macros into separate file

Will be used by the instruction compaction unit test.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/compiler: Use ARRAY_SIZE()

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2635>

intel/fs: Don't emit control barrier if only one thread is used

When there's only one hardware thread (i.e. the dispatch width greater
or equal to the workgroup size), there's no need to use a barrier to
ensure all the invocations reach the same point in the shader, because
they are already running lock-step.

Results for SKL running Iris for shader-db tests with compute shaders

    total sends in shared programs: 18361 -> 18339 (-0.12%)
    sends in affected programs: 904 -> 882 (-2.43%)
    helped: 9
    HURT: 0
    helped stats (abs) min: 1 max: 5 x̄: 2.44 x̃: 2
    helped stats (rel) min: 0.84% max: 21.43% x̄: 7.82% x̃: 2.67%
    95% mean confidence interval for sends value: -3.31 -1.58
    95% mean confidence interval for sends %-change: -14.67% -0.97%
    Sends are helped.

Shaders from Aztec Ruins, Car Chase, Manhattan and DeusEx are helped.

Results for ICL and TGL are similar to SKL.

Results for BDW are similar to SKL except for DeusEx shader that has a
workgroup size 16 but in BDW picks the SIMD8.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3226>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3226>

intel/fs: Don't emit fence for shared memory if only one thread is used

When there's only one hardware thread (i.e. the dispatch width greater
or equal to the workgroup size), there's no need to synchronize shared
memory access (SLM) since all the requests from a single thread are
already synchronized.  In such case, we just add a scheduling fence.

To be able to identify that case for all platforms, move the handling
of platforms prior to Gen11 (which don't have a separate SLM fence)
after the optimization.

Results for SKL running Iris for shader-db tests with compute shaders

    total sends in shared programs: 18395 -> 18361 (-0.18%)
    sends in affected programs: 938 -> 904 (-3.62%)
    helped: 9
    HURT: 0
    helped stats (abs) min: 1 max: 5 x̄: 3.78 x̃: 4
    helped stats (rel) min: 1.56% max: 26.32% x̄: 10.33% x̃: 2.60%
    95% mean confidence interval for sends value: -4.85 -2.71
    95% mean confidence interval for sends %-change: -19.12% -1.54%
    Sends are helped.

Shaders from Aztec Ruins, Car Chase, Manhattan and DeusEx are helped.

Results for ICL and TGL are similar to SKL.

Results for BDW are similar to SKL except for DeusEx shader that has a
workgroup size 16 but in BDW picks the SIMD8.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3226>