These are RAII-based helpers for ensuring that logic is executed, regardless of the control flow path taken.

Bug: b/139010488
Change-Id: I7bea0e550aaccf7504221a661ed2cd04e46018f9
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34769Tested-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

This small CL allows one to build and run the VulkanUnitTests
suite for Fuchsia using the Chromium build system.

Note that the output library generated by the BUILD.gn file
is also renamed libvk_swiftshader.so to correspond to the name
used by other build systems (e.g. CMakeList.txt and Android.{mk,bp}).

This also avoids a Fuchsia-specific issue, where the packaging
scripts under //build/fuchsia/ (in the Chromium src/ tree), get
confused when a generated library has the same name as one of
the system libraries (in this case, libvulkan.so, which is the
Vulkan loader provided by the SDK, and which must be copied into
the final Fuchsia package along the ICD).

Note also that most tests crash inside the generated coroutine
machine code :-/

Instructions to properly rebuild and test this:

  - Have depot_tools in your path, and a valid Chromium source
    checkout.

  - Ensure that your .gclient file contains 'fuchsia' in its
    `target_os` list variable definition, otherwise add it
    and run `gclient sync` again to fetch the Fuchsia SDK
    and other dependencies.

  - Create an output directory configured for a Fuchsia/x64 build, then
    rebuild the test suite target, e.g.:

       gn gen out/Fuchsia --args 'target_os="fuchsia" target_cpu="x64" \
           use_goma = true \
           root_extra_deps = [ \
              "//third_party/swiftshader/tests/VulkanUnitTests:swiftshader_vulkan_unittests" \
	   ]'
       ninja -C out/Fuchsia swiftshader_vulkan_unittests

  - Run the test suite with the wrapper script:

       out/Fuchsia/bin/run_swiftshader_vulkan_unittests

    NOTE: This will start a QEMU instance running a Fuchsia system image
          automatically.

+ Define VK_USE_PLATFORM_FUCHSIA when building
  for Fuchsia. This will be used by future CLs that add
  support for Fuchsia-specific extensions.

Change-Id: I687f51e98ee3cb16f6d585f44f4bb5f0631b1cd8
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34911Tested-by: David Turner <digit@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Bug: b/139010488
Change-Id: I9ea07d945a9dae6f9b92c283be0052ee68865be2
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34809Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Bug: b/139010488
Change-Id: Ie5f753df22a55dd2718b0e6ee24e48678d6a956c
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34808
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

We had these already for UInt4.

Bug: b/139010488
Change-Id: Idb9451db1221da7b309b64ef407abcea33fea21e
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34768
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

There was lots of duplicate code here.

Add SpirvRoutine::setImmutableInputBuiltins() to de-duplicate the setting of compile-time constant values.
Add SpirvRoutine::setInputBuiltin() to unify the method for setting input builtins.

Change-Id: If3c2632b4fdd74f050d2dc1494349ca359ea92fb
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35108
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Implements the following opcodes:
• OpGroupNonUniformShuffleUp
• OpGroupNonUniformShuffleDown

Bug: b/133510501
Tests: dEQP-VK.subgroups.*
Change-Id: Ia0c8b3f90b9247e7052c287601e110107dbaa17e
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35090
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Implements the following opcodes:
• OpGroupNonUniformShuffle
• OpGroupNonUniformShuffleXor

Bug: b/133510501
Tests: dEQP-VK.subgroups.*
Change-Id: I85dcfebe3d56d7c5a73b7fdbcb88885e7e228747
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35089
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Replace these with the more generic AndAll() and OrAll().
These can be used to clean up some existing code.

Bug: b/133510501
Tests: dEQP-VK.subgroups.*
Change-Id: I49f84f6c7088c1c810a0dfda5358a2cc4778f79d
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35088
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Implements the following opcodes:
• OpGroupNonUniformAll
• OpGroupNonUniformAny
• OpGroupNonUniformAllEqual

Bug: b/133510501
Tests: dEQP-VK.subgroups.*
Change-Id: Ic2ff8d99bfc7d21d7dc356eae170a95f1d016fcc
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35068
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Implements the following opcodes:
• OpGroupNonUniformBroadcast
• OpGroupNonUniformBroadcastFirst
• OpGroupNonUniformBallot
• OpGroupNonUniformInverseBallot
• OpGroupNonUniformBallotBitExtract
• OpGroupNonUniformBallotBitCount
• OpGroupNonUniformBallotFindLSB
• OpGroupNonUniformBallotFindMSB

Also don't yield for subgroup scoping - this is pointless, and does not currently work for graphics shaders.

Bug: b/133510501
Tests: dEQP-VK.subgroups.*
Change-Id: I39470bfa9f2184344d1c22e36975db0e23e48cc9
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35033
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Some capabilities affect downstream behaviour in unexpected ways. For example: https://github.com/KhronosGroup/glslang/issues/1868

Bug: b/139207336
Change-Id: I43b4f1ab3558864bb6cb93c0cbf780d0dd2e3406
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35029Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

"The CMakeSettings.json file contains information that specifies how Visual Studio should interact with CMake to build a project for a specified platform"

More information can be found here:
https://docs.microsoft.com/en-us/cpp/build/cmakesettings-reference?view=vs-2019

Bug: b/139189696
Change-Id: Ib2d6a79ad997b39538d49e17c3362818b8deaac4
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35091Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

As LValue<T> overloads operator&(), there's no sensible way to get the address of the reactor type.

This is handy for storing a pointer to an Array or LValue in a field.

Bug: b/133213304
Change-Id: I950ee312005bc57187129f009578b8715973804a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/33357Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

https://github.com/KhronosGroup/Vulkan-Headers
revision 4d49d5ced52e1ba1fe5d403726ad8291c1266d97
tag v1.1.111

Bug: b/129792032
Change-Id: Ia4e581160c4612aca44971390b31c7a497b98cfc
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/32968
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Bug: b/139184291
Change-Id: Iab77a5ac5894aec60fc70f2630b6d2a2b232d253
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/30435Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

For Vulkan we must support all blittable formats natively, so there's
no fallback to C++ code. Also, if by error an unsupported format is
used we don't have to immediately abort routine generation but can
continue with something that's benign.

This change simplifies the code, makes it more readable/maintainable,
and marks other formats as unsupported instead of unimplemented.

Bug: b/138944025
Bug: b/131243109
Change-Id: I4d246cdd280c5a352f18341996c92111a5144d96
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34588
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

R16_SNORM, R16_UNORM, and R16G16_SNORM are required for computing cube
corner pixels for decompressed EAC formats.

Bug: b/138944025
Change-Id: I35a4d448e04bba7abc0e34250d153affe7fb1db5
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34829
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

This does not yet handle rounding correctly, nor are Inf and NaN
input values dealt with.

Note that the Vulkan 1.1.117 spec refers to
https://www.khronos.org/registry/DataFormat/specs/1.2/dataformat.1.2.html#11bitfp
for the specification on the interpretation of 11- and 10-bit floating-
point formats, while that spec itself states that:
"Note that in general conversion from a real number to any
representation may require rounding, truncation and special value
management rules which are beyond the scope of a data format
specification and may be documented in APIs which generate these
formats."

The Vulkan spec however does not document any specific conversion
requirement (in contrast, it does for the shared exponent format).

Bug: b/138944025
Change-Id: If59adfb02f2793f69a028d379a03c05f9a12bd89
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34828
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

Bug: b/138944025
Change-Id: I955bb298fffe8e628a1bba73f4144598eebcdd74
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34788
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

We can avoid calling transcendental functions by manipulating the
IEEE-754 representation of the single-precision floating point values.

This implementation produces results identical to the reference code.

Bug: b/138944025
Change-Id: I95faf7e80d0604be115dd1b8dfec72f3b8ece841
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34748
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

log() was used instead of log2().

Also replace pow(2.0f, x) with the equivalent exp2(x), and remove
overloaded functions from Math.hpp which are guaranteed by C++11.

Bug: b/138944025
Change-Id: I9e87ee9b2afd8791a40dd3e73191e1ed0740acf5
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34668
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

Change-Id: I8c631f3b956d206cc1377c1d38c93f418afc20fd
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34988Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Chris Forbes <chrisforbes@google.com>
Presubmit-Ready: Chris Forbes <chrisforbes@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Without this, the ICD version is stuck at 1, and the loader refuses to
expose functions such as vkCreateWin32SurfaceKHR.

Bug: b/139184291
Bug: b/125405086
Change-Id: I8f11d1df4c5e3390fca39edaadf310de4842a3d4
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34913Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

The debugger attachment dialog uses a .rc file that was missing from the
CMake build. This caused DllMain to segfault, which the application only
observes as a failure to load the DLL.

Bug: b/139189696
Change-Id: I590c50da2410826598e8238e25f69709d80f2389
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34968
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

This is required for Cuttlefish to compile Swiftshader

Change-Id: If34d250ad9556ea753c706c739c2aed277fbaaab
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34948Tested-by: Chris Forbes <chrisforbes@google.com>
Presubmit-Ready: Chris Forbes <chrisforbes@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

This change is needed to ensure that vk_icdXXX entry points
are exported, as expected by Khronos' Vulkan loader.

Change-Id: I25b232eb3a99767072be198621bda91e3f9f9cbe
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/33050
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: David Turner <digit@google.com>
Presubmit-Ready: David Turner <digit@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

On 64-bit platforms more collisions can be prevented by spreading
the three 32-bit values more widely apart.

Bug: b/137649247
Change-Id: Icf429b4fc5a3d5d4472b346b190e48ac429bf881
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34929
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Keying off a single size_t is likely to cause collisions, more so when the hash function is simple shift-xors of 3 32-bit integers.

Use a proper cache key comparision and a better hash function while we're at it.

Bug: b/137649247
Change-Id: I1e0deefb0976102714d43dcb9dcedf6aa809bf7f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34909Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Bug: b/125909515
Change-Id: Ifc6c04959e1741d470b133b6bbe05670a7151a0f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34848Tested-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Vulkan does not have nested queries, so this was overbuilt. Get rid of
all the list wrangling and just have a single slot in the draw call for
the active occlusion query.

Bug: b/129706526
Change-Id: I30158895e6ef4edefd6e1de85766fb6c88bb528f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34728
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Bug: b/134584057
Change-Id: I1574c9017a5d48f69d14732874a8c8b814056aec
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34248Tested-by: Chris Forbes <chrisforbes@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I62deef5e2db5ccf1794f74f400cfe5d04d24f6a1
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34648
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

The last-used sampler routine can be reused when both the image and
sampler descriptors are the same. To prevent a match on uninitialized
cache keys we previously initialized them both to nullptr (as well as
the function pointer itself), at the entry of the shader routine.
This change reduces it to just initializing the image descriptor. This
suffices since all image instructions use an image descriptor
(but not all use a sampler descriptor).

Bug: b/137649247
Change-Id: I62112f12a58ddefb6e56bddab8e3fdfa76c5b756
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34628Tested-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

MakeCurrent no longer generates EGL_BAD_SURFACE when the pbuffers are
IOSurfaces. So the documentation has been updated to match.

Bug chromium:987289

Change-Id: Id074d9d6d8d0a50c91ec725ec2016a4f5ad4b6e4
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34568
Presubmit-Ready: Sean Risser <srisser@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Sean Risser <srisser@google.com>

Micro-optimization as discussed here:
https://swiftshader-review.googlesource.com/c/SwiftShader/+/34528/7/src/Pipeline/SpirvShader.hpp#1257

Change-Id: I0e71f2b5c882b2e65271a2e8f114239a924462f8
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34549Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Change-Id: I253abead4412eec7e72148c8680be04c64bd2148
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34548Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

This complements the const-cache in 34348.

Timings for the Glass demo running on a i7-4930K:

this change:               24.25 FPS
this change without 34528: 23.02 FPS
parent change (inc 34528): 22.46 FPS

Bug: b/137649247
Change-Id: I206cdaabfaf63da7f67e3cd5f6823f3343b823c8
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34528Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Re-generated Visual Studio projects using build/cmake.sh. Removed/added
changed SPIRV-Tools projects in solution.

Bug: swiftshader:132
Bug: b/123642959
Change-Id: I2a9d21a0100882d3db8ffd81d90800f464d7419b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34508Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

We previously set this device limit to infinity to avoid having to
clamp the sum of the sampler and shader provided bias values to the
[-maxSamplerLodBias, maxSamplerLodBias] range. Unfortunately infinity
is not supported/allowed by JSON and the android.graphics.cts.VulkanFeaturesTest
tests which parse the output of the vkjson tool produce a failure.

This change works around it by setting maxSamplerLodBias to a finite
value (15, which is very common according to https://vulkan.gpuinfo.org/),
and performing the clamping required by Vulkan 1.1.117 section 15.6.7.
Fortunately this cost only has to be incurred when an optional Bias
operand is used in the SPIR-V image sampling instruction. When an Lod
operand is used, which is mutually exclusive with Bias because the
former is for explicit-lod instructions and the latter for implicit-lod,
no clamping is required because the explicit LOD is not part of this
clamping equation and the sampler LOD bias must already be clamped as
specified by the spec: "The absolute value of mipLodBias must be less
than or equal to VkPhysicalDeviceLimits::maxSamplerLodBias".

Bug: b/138670766
Change-Id: Iebd6a1fef1c993ec86ac8cc38f8d9a992ba9bc47
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/34510Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>