A few functions:
- vkGetImageSparseMemoryRequirements2
- vkGetPhysicalDeviceQueueFamilyProperties2
- vkGetPhysicalDeviceSparseImageFormatProperties2

Can query properties/requirements from arrays of structures.
The arrays of structure look like:
struct Struct2{
..
Struct1 ...
...
};

We were wrongly converting them directly into arrays of Struct1
by taking the address of the Struct1 member in the first Struct2
object, which is incorrect. This change copies the logic for
vkGet* functions to vkGet*2 functions, adding a function to
directly support VkQueueFamilyProperties2.

Bug: b/142740524
Change-Id: Id1363f0ab82539d1bc1e3977c541602df8a618cb
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37276
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

* Fixes crash when window is resized.
* Take destination pitch into account when updating the framebuffer, so
we no longer get skewed renders.
* Return VK_ERROR_OUT_OF_DATE_KHR when presenting if the swapchain image
dimensions do not match the window surface's.

Bug: b/139372840
Change-Id: I744adbf222a1d06bc9cc879e299b11b56f6dd7cf
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37429
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>

Update Vulkan headers to ToT

Includes VK_EXT_line_rasterization

Bug: b/139800520
Change-Id: I054c32a29bc740679b4286818e2bac70701ef07c
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37408
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Bug: b/142132927
Change-Id: I1221f85a6a84c8ff54d900b8acb444764f18cf40
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37274Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

In addProlog, most arguments are immediately assigned to a register, for
which we emit the proper frame pointer-based move instruction. However,
for arguments that are not immediately assigned to a register, which
happens when many arguments are passed in (e.g. 10), move lowering
happens later during emitIAS, at which point, we call
getFrameOrStackReg to get the base register. The problem is that
this function will always return the stack register on Windows x86,
because needsStackPointerAlignment() is true for this target. As such,
argument access uses the stack pointer, rather than the frame pointer,
which mismatches the offsets computed for these variables in addProlog.
To fix this, stackVarToAsmOperand makes sure to use the frame pointer
for variables that are arguments (not locals) when
needsStackPointerAlignment() is true.

Bug: b/142132927
Change-Id: Ia86680381e48b61e89310f1911e72517bd14b98f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37273Reviewed-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Bug: b/142132927
Change-Id: If3d63a8327ef8940dbeee7b49c1099adda7feab2
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37275Tested-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

This change addresses the following issues:
* Microsoft x64 ABI assigns registers to the first four arguments by
argument position, not by type count.
* Microsoft x64 ABI expects caller to allocate space to copy 4 register
arguments to stack, called the Shadow Store or Home Space.
* Fix bug where preserved register area size was not computed correctly
when Xmm registers were being preserved, as it was assuming all
preserved registers were 8 bytes large.

Bug: b/142132927
Change-Id: Ibc2d82ab117c062eed2e7f66109c9d6bbdc09a8b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37272Reviewed-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>

This helps to keep the current formatting used by the Subzero codebase.

Bug: b/142132927
Change-Id: I8cb37860cf535e5b0c059316e66c19ddbafafc90
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37271Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Tests calling an external function that, itself, calls back into a
Reactor-generated routine.

Bug: b/142132927
Change-Id: I5d89000d4c23804b211aec67d5b7a36778245eba
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37270Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Bug: b/142132927
Change-Id: I2b375c48e67f026f1490e7faf3ef78df4f64693d
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37269
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>

For example, the following would fail to compile because Call tries to
return a result:

struct Class
{
    static void Foo()
    {
    }
};

Function<Void()> function;
{
    Call(SomeClass::Foo);
    Return();
}

Bug: b/142132927
Change-Id: I249144aabae1dcfc4cb04a174ae698445be382d2
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37268
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>

It seems like the sdk_version partially controls which
libc++ is used during compilation. This causes mismatches
when some targets specify a version and some targets do not.

Bug: b/138239233
Bug: b/142276229
Test: built locally
Change-Id: Idd4ede2fd3c162dd0edd850155daff70b2c05bd2
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37168
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Jason Macnak <natsu@google.com>

This change implements getHostCPUFeatures function to enable this functionality on Windows on ARM64.
This was committed to llvm at https://reviews.llvm.org/D68139 .

Bug: chromium:893460
Change-Id: I7ae62f298683979a4ab1aa8099d6e91706b15c9b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37328
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

We were previously checking if we could read to the end of the last
row in a pixel buffer object without overrunning the buffer. Instead
we need to check if we could read to the last byte defined by the
parameters and state without overrunning the buffer.

Also added a unittest to ensure that we don't reject valid PBO
operations, and do reject invalid PBO operations.

Bug: 140881221
Change-Id: I9b8879f1be30ec4fe6d60fe43f8ff4078f215868
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37228
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Sean Risser <srisser@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

It looks like clang-cl is clobbering a floating point register
when passing a floating point argument to a DLL, because the
first time glUniform1f() is called, the value is set to 0.0f.
All subsequent glUniform1f() calls work properly.

I changed glUniform1f() calls to glUniform1fv() calls in the
AtanCornerCases test to get the test to pass again, but this
is just masking the compiler issue.

Bug: chromium:1013375
Change-Id: I8363182f26c50cd50c07ccea65ab2eb640a7da76
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37348
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Change-Id: Ie1a4d1a9d562486c8132c0c8f934b196a57f7459
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37308
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Chris Forbes <chrisforbes@google.com>

https://swiftshader-review.googlesource.com/c/SwiftShader/+/35939
was submitted without being properly rebased. Unfortunately, the
merge was successful on the gerrit server, but led to a broken
Linux build for swiftshader_libvulkan. This fixes the issue.

Bug: 140419396
Change-Id: Ib3167e11ea4c4957cf28fa7b45305e98ca27c413
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37289
Kokoro-Presubmit: David Turner <digit@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: David Turner <digit@google.com>
Presubmit-Ready: David Turner <digit@google.com>

Add external semaphore support for Linux and Android
based on a process-shared pthread mutex + condition
variable stored in a shared memory region backed by
memfd_create().

This takes care of waiting for external semaphores
in a background thread when invoked from a Yarn fiber.

Test: dEQP-VK.api.external.semaphore.opaque_fd*

Bug: b/140421726
Change-Id: Ifa74c807d3e33914e5a37dd96650c312246c8e4f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35939Reviewed-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: David Turner <digit@google.com>
Kokoro-Presubmit: David Turner <digit@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

'linux' turns out to be a prebuilt macro for GCC (not Clang)
so using "namespace linux { ... }" actually does not compile.

This CL fixes the issue by removing the namespace, and using
Linux as a prefix for the corresponding class.

Bug: 140419396
Change-Id: I1e980bc66b2cf8f2e21d9f7076918e2b476ae4d4
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37288
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: David Turner <digit@google.com>
Presubmit-Ready: David Turner <digit@google.com>

In order to perform a blitFramebuffer operation, ANGLE uses
a compute shader which reads the stencil from a depth+stencil
image to a buffer, and then copies this buffer to the stencil
of the other image.

This cl basically treats VK_FORMAT_S8_UINT as VK_FORMAT_R8_UINT
with quad layout for the purpose of texture sampling.

Fixes the following tests:
dEQP-GLES3.functional.fbo.blit.depth_stencil.depth32f_stencil8_basic
dEQP-GLES3.functional.fbo.blit.depth_stencil.depth32f_stencil8_stencil_only
dEQP-GLES3.functional.fbo.blit.depth_stencil.depth24_stencil8_basic
dEQP-GLES3.functional.fbo.blit.depth_stencil.depth24_stencil8_stencil_only

Bug: b/142385547
Change-Id: Ib2ea7fa81496ceca1c24ea7f065b1c2cd05596ee
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37188
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

Bug: b/141676114
Change-Id: I47e7d90e14b44533e64d352ecc6440495c0b7d3f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36597
Presubmit-Ready: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>

Bug: b/141380498
Change-Id: I25c6bf769c558fb4bc3a3a64f66289cf5dbfb879
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36595Tested-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Mostly copypasta from: src/Reactor/ExecutableMemory.cpp

Will be upstreamed to marl if the autorollers are happy.

Bug: b/140546382
Change-Id: I6cf0dc8638e37f17d097ea645c9a84cef1368a3d
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37131Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

These build files are not upstreamed to the marl repo as they use SwiftShader specifics.
This means the file sets need to be updated whenever marl is updated.

Bug: b/140546382
Change-Id: I18181d95d2c8162525afc4fe7d49106bc2186562
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37130Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

* changes:
  Update Marl to 49e4e31
  Squashed 'third_party/marl/' changes from 12872a0df..49e4e3141

Bug: b/141863676
Change-Id: Ie6ea68edcc0505bf2931c0fe91e4be418c4a21e0
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37108Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Includes fix for build failure in marl/blockingcall.h on old versions of GCC.

Changes:
  49e4e3141 Add blocking_call test for functions with non-void return.
  0acd85c63 blocking_call: Workaround for GCC bug with parameter packs in lambdas.
  ce47eca32 Work around TSAN false-positive in old versions of GCC
  85854b73a Implement allocation page guards
  d8b38213f Add missing header include guards
  776f7a485 Add Allocator interface and use them throughout marl.

Commands:
  git subtree pull --prefix third_party/marl https://github.com/google/marl master --squash

Bug: b/140546382
Change-Id: I2b52f28b373bd724912ba13a208f1719e80a3638

49e4e3141 Add blocking_call test for functions with non-void return.
0acd85c63 blocking_call: Workaround for GCC bug with parameter packs in lambdas.
ce47eca32 Work around TSAN false-positive in old versions of GCC
85854b73a Implement allocation page guards
d8b38213f Add missing header include guards
776f7a485 Add Allocator interface and use them throughout marl.

git-subtree-dir: third_party/marl
git-subtree-split: 49e4e314157e8560f6ad620f5ec28a877612f03b

Bug: b/135744933
Change-Id: I6b37091c6390e5c3c67a608b85ecd3d9aef07580
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37028
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>

CMake (without or without Visual Studio 2019) should be used instead
for the Vulkan build.

Bug: b/139189696
Change-Id: Ib87e34e2a210944446b8ea7bd57504acc4bc98d7
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37068
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

LLVM 8 removes several x86 SSE intrinsics for saturated arithmetic, but
provides target-independent replacements. Note there are now three code
paths for these vector operations:
- With LLVM 8+, use the new target-independent intrinsics.
- With LLVM 7 on x86, use the SSE intrinsics.
- On other architectures, lower to a generic sequence of vector
  operations (lowerPSAT).

Bug: b/139412871
Change-Id: I849f8b5c004849c628507d6377779df170a11d93
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36788
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Hold a reference to the scheduler for each vulkan device.

This attempts to work around Windows-specific teardown issues, possibly caused by dependency atexit() logic.

Bug: b/141380274
Change-Id: I5a34c18f8e3738f02221f1784419bc21a6430067
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36990
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

"set -e" was aborting the script after grep doesn't find the Bug line,
and "set -x" makes things overly verbose. So perform the validation
before these options are set.

Also return 1 instead of 0 on failure.

Bug: b/141892461
Change-Id: Ica0364ff49383b47cf37b3a02d6b03e55d8d88ad
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37008
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

Proper implementation of a VkSemaphore using an std::mutex
and a marl::ConditionVariable. This should make the implementation
compatible with both fibers and threads at the same time.

A future CL will add platform-specific implementations to cover
external semaphores as well. Which explains why the implementation
details are hidden from VkSemaphore.hpp.

Bug: b/140421726
Change-Id: I1db55493a41db0eb60ce9181fe864253db09f4f8
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35933Tested-by: David Turner <digit@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

GLsizei is historically defined as a 32-bit signed integer, which is
not allowed to be negative. But doing arithmetic with them can lead to
overflows producing undefined values, including negative ones.

Use size_t instead for these image data size calculations. It's defined
as an unsigned integer large enough to fit valid memory buffer sizes.
In practice that's 32-bit on 32-bit platforms, and 64-bit on 64-bit
ones, which is large enough to avoid overflow in the OutOfMemory test.

Bug: b/140421004
Change-Id: Id05dd64d470f579455fdc9925b0ab303ecaf37fb
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36968Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

The Git change's commit message should have a line that starts with Bug:
or Issue:, followed by a bug ID. This is important for tracking purposes,
association with follow-up changes, and having a centralized place to
discuss issues.

Note that this change doesn't actuall check that the bug ID is valid.
Also, at this point Kokoro doesn't block submits.

Bug: b/141892461
Change-Id: Idf57534d24842e18dca90ec01e5bb134db612286
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36868
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Not tested by dEQP (and unlikely to be, as we don't want to force
drivers to break existing content by adding new negative coverage in
general) but this was wrong.

Bug: b/141916742
Test: dEQP-EGL.functional.image.api.create_image_*
Change-Id: I774d197a9c5c47447c97d212b41e4c33ad963474
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36949Tested-by: Chris Forbes <chrisforbes@google.com>
Presubmit-Ready: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

We need the face target here, not the whole texture target.
This fixes the one case that was actually broken, and avoids hitting
an assert in debug swiftshader for all tests in this group.

Bug: b/141916742
Test: dEQP-EGL.functional.image.api.create_image_gles2_cubemap*
Change-Id: I25ef8f6d6e2909adfc646e68f1711589777017ba
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36948Tested-by: Chris Forbes <chrisforbes@google.com>
Presubmit-Ready: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Implemented operator< with std::tie to force strict weak ordering.

Bug b/141863676

Change-Id: If14328922f0c8c62fd92d74f1bcca0c9a2e77f4c
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36929
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Two issues here:
- Logic was inverted
- getTopLevel() isn't quite right. The condition in the spec is
  whether there are any nonzero levels defined; it does not require them to
  be contiguous with the base level.

Test: dEQP-EGL.functional.image.api.create_image_gles2_tex2d_rgba_level0_only
Bug: b/141916742
Change-Id: I336d628c9508bfe61d1471431cc6170454a24d1b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36888Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Chris Forbes <chrisforbes@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>