#80757 has been merged.

The two types duplicated mostly the same values.
Here they are decomposed to carry orthogonal and complimentary
information.

Use `utils::IteratorType` instead of `mesh::IteratorType`. It now has
only parallel and reduction values.

Rename `Partial` to `ReductionKind`.

Add `getReductionLoopIteratorKinds` method to `ShardingInterface`.

Add test case for store suppression that still trigger after
https://github.com/llvm/llvm-project/pull/81749

- Add a single `cmp` function to derive all comparison operators
 - Use the `friend` version of the member functions for symmetry
 - Add a `is_neg` function to factor sign extraction
 - Implement binary op through macro expansion

This caused links to fail with:

  lld/MachO/Symbols.cpp:97:
  virtual uint64_t lld::macho::Defined::getVA() const:
  Assertion `target->usesThunks()' failed.

or crash when asserts are disabled. See comment on
https://github.com/llvm/llvm-project/pull/79894

> Enable chained fixups in lld when all platform and version criteria are
> met. This is an attempt at simplifying the logic used in ld 907:
>
> https://github.com/apple-oss-distributions/ld64/blob/93d74eafc37c0558b4ffb88a8bc15c17bed44a20/src/ld/Options.cpp#L5458-L5549
>
> Some changes were made to simplify the logic:
> - only enable chained fixups for macOS from 13.0 to avoid the arch check
> - only enable chained fixups for iphonesimulator from 16.0 to avoid the
> arch check
> - don't enable chained fixups for not specifically listed platforms
> - don't enable chained fixups for arm64_32

This reverts commit 775c2856.

If a split memory access introduced by SROA accesses precisely a single
field of the original operation's !tbaa.struct, use the !tbaa tag for
the accessed field directly instead of the full !tbaa.struct.

InstCombine already had a similar logic.

Motivation for this and follow-on patches is to improve codegen for
libc++, where using memcpy limits optimizations, like vectorization for
code iteration over std::vector<std::complex<float>>:
https://godbolt.org/z/f3vqYos3c

Depends on https://github.com/llvm/llvm-project/pull/81285.

We need to always emit the diagnostic, but still do the cast.
This is also what the current interpreter does.

Update truncateToMinimalBitwidths to handle truncating ICMPs. For ICMPs,
the new target type will be the same as the original type. In that case,
only truncate the operands, but skip the extend. This is in line with
what the original truncateToMinimalBitwidths did for compares.

Fixes https://github.com/llvm/llvm-project/issues/81415.

When the data-in-code entries are in separate sections, they are not
guaranteed to be sorted. In particular, 68b1cc36f3df marked some libc++
string functions as noinline, which leads to global ctors involving
strings now producing data-in-code sections in __TEXT,__StaticInit,
which is why this now happens in practice.

Since data-in-code entries are relatively rare and small, just sort
them. No observed performance impact.

See also crbug.com/41487860

We support '_Alignas' from C11 as an extension in C++. However, we were
not correctly parsing its use in local variable declarations. This patch
addresses that issue.

- Allow/force to break the line or not.
- Allow/force to insert space or not.

[bazel] Port for 0eedc85b and 82a4a416

While we _do_ get them registered via visitInitializer(), they
are still local, so gone on the next call to e.g. evaluateAsRValue().

Visit them lazily, similarly like we do in C.

This adds ops for the two and four-way SME 2 multi-vector zips.

See:

-
https://developer.arm.com/documentation/ddi0602/2023-12/SME-Instructions/ZIP--two-registers---Interleave-elements-from-two-vectors-?lang=en
-
https://developer.arm.com/documentation/ddi0602/2023-12/SME-Instructions/ZIP--four-registers---Interleave-elements-from-four-vectors-?lang=en

This needs to be done but we didn't use to do it.

We have no need to vectorize affine.apply inside the vectorizing loop.
However, we still need to generate it in the original scalar form. We
have to replace all its operands with the generated scalar operands in
the vectorizing loop, e.g., induction variables.

[Transforms][Utils] Add helpers to map between Reduction IntrinsicID and Arithmetic Instruction Opcode and MinMax IntrinsicID / RecurKind

Noticed on #81852

Found when compiling openssl master branch using clang-cl.

This commit introduces usage of InterlockedAdd64:

https://github.com/openssl/openssl/commit/d0e1a0ae701cfaca7f3dd3bf28a3f934a6408813


https://learn.microsoft.com/en-us/cpp/intrinsics/interlockedadd-intrinsic-functions

This fixes a large vertical blank space at the top of the page that was caused
by the floating sidebar.

Bring list of supported formats in docs back in sync with the code.

Just like we did with the l-to-r conversion, we need to do this
while the data is still alive.

This is also necessary for enabling ClangBuiltLinux:
https://github.com/ClangBuiltLinux/linux/issues/1530

As detailed in [GCC PR
sanitizer/113785](https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113785),
the GCC test `c-c++-common/asan/swapcontext-test-1.c` `FAIL`s on
Solaris/sparc.

This is due to the fact that Solaris 10/SPARC changed the semantics of
`makecontext` so `ucontext_t.uc_stack.ss_sp` refers to the stack base
address. To maintain binary compatiblity, the external name was changed
to `__makecontext_v2`, keeping the old version.

To match this, `__makecontext_v2` needs to be intercepted instead of
`makecontext`.

Tested on GCC trunk on `sparc-sun-solaris2.11`, `i386-pc-solaris2.11`,
and `x86_64-pc-linux-gnu`.

Also tested on the same targets on LLVM `main`. However, this only
proves that Linux/x86_64 isn't broken, since all `makecontext` tests are
Linux-specific.

Retain TBAAStruct if we fail to match the access to a single field. All
users at the moment use this when using the full size of the original
access. SROA also retains the original TBAAStruct when accessing parts
at offset 0.

Motivation for this and follow-on patches is to improve codegen for
libc++, where using memcpy limits optimizations, like vectorization for
code iteration over std::vector<std::complex<float>>:
https://godbolt.org/z/f3vqYos3c

Depends on https://github.com/llvm/llvm-project/pull/81284

It slipped in from commit fa6025e2.

This was done inconsistently before. Many instructions used the default
"AMDGPU" namespace which I would like to remove.

This commit fixes memory leaks in sparse tensor integration tests by
adding `bufferization.dealloc_tensor` ops.

Note: Buffer deallocation will be automated in the future with the
ownership-based buffer deallocation pass, making `dealloc_tensor`
obsolete (only codegen path, not when using the runtime library).

The setting name that was settled on is symbols.auto-download.

Descriptor addendum have a field to hold length parameters (currently
only one). This field is currently never used because flang does not
lowered derived types with length parameters.

However, leaving it uninitialized is causing bugs in code like gFTL
where the code is trying to sort POINTERs (see [1]). More precisely, it
is an issue when two pointers should compare equal (same base address),
because the uninitialized values in the addendum may differ depending on
the "stack history" and optimization level.

Always initialized the length parameters field in the addendum to zero.

[1]:
https://github.com/Goddard-Fortran-Ecosystem/gFTL/blob/dc93a5fc2fe2b487590580d9f0a17d77ca3fe5c1/include/v1/templates/set_impl.inc#L312

The type being transferred to an integer array may look like:

```
   TYPE :: localwrapper
    TYPE(T), POINTER :: item
   END TYPE localwrapper
```

Which in flang case ends-up tra...

A class is added that contains common functions and data members that are used in many of the "eval" functions. This results in shorter "eval" functions and less code repetition.

This patch implements the codegen support of zabha (Byte and Halfword
Atomic Memory Operations) v1.0-rc1 extension.
See also https://github.com/riscv/riscv-zabha/blob/v1.0-rc1/zabha.adoc

.

---------

Co-authored-by: Craig Topper <craig.topper@sifive.com>

This fixes https://github.com/llvm/llvm-project/issues/81555

This is to avoid warnings when invoked from the flang documentation
generation build.

The warning can be seen in the CI
(https://lab.llvm.org/buildbot/#/builders/89/builds/57451).
```
/home/buildbot/as-worker-4/publish-sphinx-docs/llvm-project/mlir/tools/mlir-tblgen/OpDefinitionsGen.cpp:2956:38: warning: unused variable ‘operand’ [-Wunused-variable]
```

Just forward the pointer.
Copy tests from SemaCXX