This is the same fix that I applied to LLVM7 here:
https://swiftshader-review.googlesource.com/c/SwiftShader/+/37608
but this time to subzero-llvm.

Fixes ANGLE dEQP tests run against SwiftShader/Subzero.

Bug: chromium:151653536
Change-Id: Ia4725b74d068a3afd99d407c53b11f757bc5e11f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42469
Presubmit-Ready: Antonio Maiorano <amaiorano@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

This CL refactors the implementation of VkSemaphore objects
in the following way:

- Add the ability to support several external handle types
  concurrently. Before this CL, each platform could support
  a single handle type (e.g. on Linux,
  VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT only).

  The changes here will allow future CLs to support more than
  one type per platform (e.g. the Linux implementation may
  support OPAQUE_FD_BIT as well as SYNC_FD_BIT at the same
  time).

- Better implementation of temporary imports. In particular,
  the following sequence now works properly:

    1) Create exportable semaphore A.
    2) Export A to an external handle/descriptor.
    3) Signal A.
    4) Temporarily import _another_ handle into A.
    5) A.wait()   // waits on the temporary payload, then discard it.

  Before the CL, A would end up, incorrectly, unsignalled. Because
  the export operation created an External instance that held
  the payload modified in 3), which was then discarded after
  the wait() in 5).

- Improved and consistent handling of errors during import/export
  operations, through the use of templates.

+ Add a technical note in VkSemaphore.h explaining how everything
  works, since there are several subtle points in the spec.

Bug: b/140421736

Change-Id: I9b6935db3238fec7af8e0c81666e2f5c72075756
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39880Tested-by: David Turner <digit@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Kokoro-Presubmit: David Turner <digit@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Change-Id: I2833d77b233998725288f4df3ddee5138805981d
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42349Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Previously nested scopes would hide their parent variables.
This is no longer the case.

Bug: b/148401179
Change-Id: Icf2771a525524b8796ab12e295e141388bb5b399
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42368
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

This allows Chromium to use the SwiftShader Vulkan library
with the Subzero backend when available.

Bug: chromium:1060139
Change-Id: I444b3e6844c52781ada9f51c8ddb4e4d2ea6a6d4
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42108
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

... between the `OpFunction` and `OpLabel`.

It is legal to have `OpLine`s between here, which we were incorrectly warning about.
The `OpenCL.Debug.Info` opcodes can also sit here, which are numerous. Just silence the warning, we're only scanning forward for the `OpLabel` here.

Bug: b/148401179
Change-Id: Ia9b8b024a1a15bbf431abec8f8ec47e2b735ed1b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42328Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Subzero maintains an internal scheduler that it uses only if there's no
external scheduler already bound on the current thread. This removes the
dependency on Marl in ReactorUnitTests, as we can rely on the internal
one being used.

Bug: b/145754674
Change-Id: Iddbaa299ccd904a87b8aa86e82e5517c9d72ef59
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41789Tested-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

To fix broken Chromium build:
https://chromium-review.googlesource.com/c/chromium/src/+/2102747

Bug: fix Chromium build
Change-Id: I00e3bbe3087d6c885164d6b91e55ac50a89b5dad
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42348Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>

Instead of ON/OFF and 1/0. TRUE/FALSE are the original Boolean classes:
https://cmake.org/pipermail/cmake/2007-December/018548.html

Bug: b/145758253
Change-Id: Ida6a0d81da37914a01d7da9ecf2bb2d2f840e369
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42148Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

This ensures these intermediate targets only get built when needed,
and allows reducing the complexity of conditional logic.

Bug: b/151250656
Change-Id: I3cbc2cc7a4c3d6f3691eeeef55d381ef03c2db9f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42308
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

* This change was authored by bclayton@, with some modifications.
* Replaces Win32 fiber implementation with marl tasks, making coroutines
  work on all marl-supported platforms.

Bug: b/145754674
Change-Id: Ic3de82afc69549e1d56688c6faf8077a6f446ee0
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41788
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>

The `PRINT_EACH_EXECUTED_INSTRUCTION` preprocessor flag complements the existing `PRINT_EACH_PROCESSED_INSTRUCTION` flag to print each instruction before they are executed.
This is exceptionally helpful when debugging the debugger.

Bug: b/148401179
Change-Id: If35728f115d664c1dba63e625eefccac9d3140ce
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42249
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Reading between the lines of the spec, Functions are also scopes.

Move common scope fields into the base Scope class.

Fix `scope` fields requiring a `LexicalBlock` to take the more general `Scope`.

Bug: b/148401179
Change-Id: Iac7b4fd4440a701905e2dbcdf71b91146ea8ff7d
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42248Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

Makes for easier reading of the assertion macros.

Bug: b/148401179
Change-Id: Icc48baa9c476967c6dea19c2f0e91ed253dce323
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42194Tested-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Acts as a placeholder for no-debug-info. Implemented as nullptr.

Bug: b/148401179
Change-Id: I6cbfb227296d98befa6a7d2baa29128cb0c7bd29
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42193
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Bug: b/148401179
Change-Id: I3a5cacc24bcb31a0b7e7cc3319ba500de59baa8a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42192Tested-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

The inlined-at isn't currently used, but this unblocks tests that emit this instruction.

Bug: b/145351270
Change-Id: I787e0de3d7ede8225ad3a86544f867559ea9c387
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42095
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

According tot he Vulkan spec, about vkCmdBlitImage:
"As the sizes of the source and destination extents can
 differ in any dimension, texels in the source extent
 are scaled and filtered to the destination extent."

Logic was added to allow filtering in 3 dimensions in
the Blitter. Since it requires more instructions, a
state member was added in order to only filter in Z
whenever we actually to do so (when the number of
slices differ between source and destination regions).

Fixes all tests mentioned in the issue.

Bug: b/150155499
Change-Id: I63da92db94d8d0394a06271cd47d4930cf5f726a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42208
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>

Bug: b/151314247
Change-Id: Ib10831c42a0c18f86397ed3b6c5d32a578653c89
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42288
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>

We currently rely on limiting the mipmap level count of YCbCr images
and the presence of sampler YCbCr conversion to disable mipmapping.

This is valid, but fragile. The Vulkan 1.2 spec guarantees that when
sampler Y'CbCr conversion is used, the image view must have a Y'CbCr
format:

"If the descriptor refers to a sampler that performs Y'CbCr conversion,
 the sampler must only be used to sample the image in the same
 descriptor."

"If the image view is to be used with a sampler which supports sampler
 Y'CbCr conversion, an identically defined object of type
 VkSamplerYcbcrConversion to that used to create the sampler must be
 passed to vkCreateImageView in a VkSamplerYcbcrConversionInfo included
 in the pNext chain of VkImageViewCreateInfo. Conversely, if a
 VkSamplerYcbcrConversion object is passed to vkCreateImageView, an
 identically defined VkSamplerYcbcrConversion object must be used when
 sampling the image."

"maxMipLevels is the maximum number of mipmap levels. maxMipLevels must
 be equal to the number of levels in the complete mipmap chain based on
 the maxExtent.width, maxExtent.height, and maxExtent.depth, except when
 one of the following conditions is true, in which case it may instead
 be 1:
  * ...
  * image format is one of those listed in Formats requiring sampler
    Y′CBCR conversion for VK_IMAGE_ASPECT_COLOR_BIT image views"

Bug: b/151263485
Change-Id: Ibb8e3fe34ecda04d6fb41308437fa4c66815f1fb
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42228
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Bug: b/149537975
Change-Id: Ia589e0015437e5257991fe2e8b5f6459384a76de
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42048
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

SetupRoutine uses V0 == V1 == V2 when drawing points, so we don't have
to duplicate those vertices explicitly.

Bug: b/149537975
Change-Id: I723a7139f5eb5735bedb0c17796c00dc4b414fa7
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42028
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

If the Reactor backend is Subzero, there's no need to build LLVM. This
is accomplished by excluding the LLVM target from the "all" target.

Bug: b/151250656
Change-Id: I26dc44921cc44d379c619700c6dd03f6fe94ee27
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42229
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>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

A single change to fix windows breakage due to 'alloc' requiring an underscore prefix on some toolchains.

Changes:
    64d123947 thread.cpp: Replace use of alloca with vector.

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

Bug: b/140546382
Bug: angleproject:4474
Change-Id: If15baa2531a5b49537bb0ba9f190454f2044ad9b

64d123947 thread.cpp: Replace use of alloca with vector.

git-subtree-dir: third_party/marl
git-subtree-split: 64d123947e4dbc741217ea7b49f3d2b6275be117

Also contains the following changes:
* Replace the use of `unsigned char` with `uint8_t`. `char` is not guaranteed to be 8-bit.
* Remove the `dstW` and `dstH` parameters from `BC_Decoder::Decode` and `ETC_Decoder::Decode`. They're always the same as the `w` and `h` parameters.
* Add BC6 types to various switch cases. The actual decode logic is not implemented for these formats.

Tests: *bc7*
Bug: b/151203718
Change-Id: I7b232b9dc3a9b02d172f87a62c88ce56b2cca956
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41508Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Contains a fix for an arm linker error (chromium:1058107) and
scaling beyond 64 threads on Windows.

Changes:
    94a361cf0 thread.cpp Fix minor bug in getProcessorGroups()
    0249a2624 Thread: Use WaitForObject for Thread::join()
    62f209bbb arm: Annotate marl_fiber_swap as type %function
    00f091e08 Update README.md
    773d9f475 Add new example 'tasks_in_tasks'
    3f69e73ce Scheduler: Replace use of std::thread with marl::Thread

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

Bug: b/140546382
Bug: chromium:1058107
Change-Id: Ic5a96ad5f054a19f6a5a77e1f106c73ba60c0a78

94a361cf0 thread.cpp Fix minor bug in getProcessorGroups()
0249a2624 Thread: Use WaitForObject for Thread::join()
62f209bbb arm: Annotate marl_fiber_swap as type %function
00f091e08 Update README.md
773d9f475 Add new example 'tasks_in_tasks'
3f69e73ce Scheduler: Replace use of std::thread with marl::Thread

git-subtree-dir: third_party/marl
git-subtree-split: 94a361cf0b5187a5845f3a37c9e67c8406ccb40b

By setting the RPATH to $ORIGIN, Linux executables can behave like on
Windows and search for shared libraries in the exectuables directory
before looking elsewhere. We copy SwiftShader's Vulkan library into this
directory, naming it vulkan.so.1 to match the loader name the
executables look for.

This also relies on SwiftShader exporting all Vulkan API entry functions
like the loader library does: b/148240133

Bug: b/141361933
Change-Id: I617303cc11faead7aac9e689142a4400731a58a5
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41970Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

Test: none
Change-Id: I10ca476ed7dade95b516fabc1279fe751285114f
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42128Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

Using --dailynow --dailychange <hash> makes Regres check out the specified
change to obtain the vk-master.txt from and perform a full run of all the
tests.

Also fix --keep usage.

Bug: none
Change-Id: Ia126bfc136e6b86a6fcc570f79a9293ae97ca392
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42109Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

https://swiftshader-review.googlesource.com/c/SwiftShader/+/40788 and
https://swiftshader-review.googlesource.com/c/SwiftShader/+/41968 cause
us to use different compile flags for Subzero's LLVM dependencies than
the rest of SwiftShader. Suppress the warnings that this has unleashed.

Bug: b/132445520
Change-Id: I1f20e32b487035e953c02cd4829460c072e38ac3
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41969
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Add `filter` flag to run_testlist so that tests can be filtered by wildcard or regex.
Update `run_testlist.sh` and `run_testlist.bat` to automatically use the vk-master.txt test list.

Change-Id: I94ef1d2e9220f18bdf50555db8c291b3736ec3f3
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42089Reviewed-by: Paul Thomson <paulthomson@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

The http URL now results in a redirect to an https URL.
Update the URL to the new redirect URL.
Also add the `-L` flag to curl so that future redirects don't break our presubmits.

Also verify with the provided PGP signatures instead of using a more vulnerable sha1.

Bug: none. Fixing presubmits.
Change-Id: I715c45f7083bf0f5528d82ba93d008464d0d7f88
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42093Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

On some ABIs, C++ functions that return bool only set the least
significant bits of the return register (e.g. AL on x64). The Call
implementation in SubzeroReactor would use uninitialized i32 target
to hold the result of a bool-returning function, but this would return
erroneous results when the called function would return false by setting
the LSB to 0, but with a non-zero MSB, thus returning true.
This change makes sure to truncate the i32 result to i8 (bool),
returning true only if the lsb is non-zero.

Bug: b/151158858
Change-Id: I41055a94d7f8045da503f27881ed887c1926f77b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42068Tested-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I57d7aaaaf85f009721b791ad15f20ea7971a9092
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41988Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

Bug: b/149537975

Tests: dEQP-VK.*point*

Change-Id: I22ee0c0ebc4e7b414e055e2b87010004dfc8741a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41948Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Marc-Antoine Desroches <madesroches@google.com>

This is a partial revert of https://swiftshader-review.googlesource.com/c/SwiftShader/+/40788
where .c files were renamed to .cpp to avoid a warning treated as error:
command line option ‘-fno-operator-names’ is valid for C++/ObjC++ but not for C [-Werror]
This appears to have caused macOS linker issues: chromium:1059196

Instead, explicitly add the flag to compile with warnings-as-errors
(when enabled) for code that is actively maintained by us, and make
Subzero's legacy LLVM dependencies a separate library compiled without
warnings-as-errors.

Also revert CMakeSettings.json changes that were inadvertently committed.

Bug: b/141361933
Bug: chromium:1059196
Change-Id: I27dd57d954a413a710613b0bfa2e69bd5242853a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41968Reviewed-by: Marc-Antoine Desroches <madesroches@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>

On Windows placing the vulkan-1.dll next to the executables suffices for
them to load SwiftShader instead of the system's Vulkan ICD loader (if
present).

Also make warnings-as-errors only affect our own projects, and put
things into project folders.

Bug: b/141361933
Change-Id: Ibdf8bcb992aa484d4bebaaf52ae77a8ced61709a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/40788
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>

Implement non-optimized version. Should generalize and optimize code in
UShort4::UShort4(RValue<Float4> cast, bool saturate) so we can use it
here.

Bug: b/145754674
Change-Id: I16a24dab2979bb97d2d5f36a406aeb62aeb976e8
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/41928Tested-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>