Changes

[SVE] Lower fixed length VECREDUCE_[FMAX|FMIN] to Scalable

Differential Revision: https://reviews.llvm.org/D88444

[flang][msvc] Define implicit conversion from UnsignedInt128 to int64_t.

The custom implementation of UnsignedInt128 has an implicit conversion operator to unit64_t, but not int64_t. Considering that the former is already truncating, and C++ implicitly converts uint64_t to int64_t, UnsignedInt128  should also support an implicit conversion to int64_t. An analogous conversion would be from uint32_t to int16_t.

Without the conversion operator overload, the msvc emits the following error:
```
descriptor-io.h(44): error C2440: 'static_cast': cannot convert from 'A' to 'int64_t'
        with
        [
            A=Fortran::common::uint128_t
        ]
```

This patch is part of the series to make flang compilable with MS Visual Studio <http://lists.llvm.org/pipermail/flang-dev/2020-July/000448.html>.

Reviewed By: klausler

Differential Revision: https://reviews.llvm.org/D88509

[flang][msvc] Define access flags under Windows. NFC.

The flags F_OK, R_OK and W_OK are defined in unistd.h, which does not exist under the Windows platform. Windows still defines the `access` function. Its access flags are documented at https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/access-waccess. For compatibility, define the flags F_OK, R_OK and W_OK using these constants.

This patch is part of the series to make flang compilable with MS Visual Studio <http://lists.llvm.org/pipermail/flang-dev/2020-July/000448.html>.

Reviewed By: klausler

Differential Revision: https://reviews.llvm.org/D88508

[nfc][libomptarget] Drop parameter to named_sync

[nfc][libomptarget] Drop parameter to named_sync

named_sync has one call site (in sync.cu) where it always passed L1_BARRIER.
Folding this into the call site and dropping the macro is a simplification.

amdgpu doesn't have ptx' bar.sync instruction. A correct implementation of
__kmpc_impl_named_sync in terms of shared memory is much easier if it can
assume that the barrier argument is this constant. Said implementation is left
for a second patch.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D88474

Improve the representation of <compare>'s zero-only type.

* Use an empty struct instead of a member pointer to represent this
  type, so that we don't actually pass a zero member pointer at runtime.

* Mark the constructor as consteval to ensure that no code is emitted
  for it whenever possible.

* Add a honeypot constructor to reject all non-int arguments, so that
  the only argument that can arrive at the real constructor is the
  literal 0.

This results in better generated code, and rejecting invalid comparisons
against nullptr, 0L, and so on, while also rejecting invalid comparisons
against (1-1) and similar that would be allowed if we required an
integer constant expression with value 0.

Differential Revision: https://reviews.llvm.org/D85051

Recognize setjmp and friends as builtins even if jmp_buf is not declared yet.

This happens in glibc's headers. It's important that we recognize these
functions so that we can mark them as returns_twice.

Differential Revision: https://reviews.llvm.org/D88518

[AMDGPU] Remove SIEncodingFamily.GFX10_B

It turns to be not needed anymore.

Differential Revision: https://reviews.llvm.org/D88520

Fix use of wrong printf format specifier for size_t argument.

This causes a build break under -Werror=format.