Accommodate operations with VALUE dummy arguments in the runtime support
for the REDUCE intrinsic function by splitting most entry points into
Reduce...Ref and Reduce...Value variants.

Further work will be needed in lowering to call the ...Value entry points.

Neither isConstant() not isConstQualified() return true for these.

Reverts llvm/llvm-project#93443

For bolt/test/runtime/X86/exceptions-pic.test, llvm-bolt seems to call
emitLabel twice and the assert will fail. Work around it after
2cc4bc13

blendvs are very similar to select, so we adjust
arguments and forward them into select handler.

This reverts the functional elements of commit
3e78fa86 and redoes it, by fixing the
true root cause of #61317.

A TemplateName can be non-canonical; profiling it based on the canonical
name would result in inconsistent preservation of as-written information
in the AST.

The true problem in #61317 is that we would not consider the methods
with requirements expression which contain DTSTs as the same in relation
to merging of declarations when importing modules.

The expressions would never match because they contained DTSTs pointing
to different redeclarations of the same class template, but since
canonicalization for them was broken, their canonical types would not
match either.

---

No changelog entry because #61317 was already claimed as fixed in
previous release.

`#` and `##` preprocessing tokens cannot be replaced by constexpr
function. It should be ignored in check.

Add an `fastMathAttr` on `arith::extf` and `arith::truncf`. If these two
ops are inserted by some promotion passes (like legalize-to-f32 /
emulate-unsupported-floats), they will be labeled as
`FastMathFlags::contract`, denoting that they can be then `eliminated by
canonicalizer`.

The `elimination` can help improve performance, while may introduce some
numerical differences.

Follow-up to a91c8398.

so that tools like llc can use the option as well.
ca91538c is a prerequisite.

Also delete `AllowTemporaryLabels = true` from MCContext::reset: when
llc supports -save-temp-labels in the next change, this assignment
should be removed to support -compile-twice.

Cortex-A725 and Cortex-X925 are Armv9.2 AArch64 CPUs.

Technical Reference Manual for Cortex-A725:
   https://developer.arm.com/documentation/107652/latest

Technical Reference Manual for Cortex-X925:
   https://developer.arm.com/documentation/102807/latest

The test has a check that the static linker supports the new option,
but it assumed the Xcode 16 linker also meant it was running on
macOS 15 and the dynamic linker would honor dependencies flagged
this way.  But Xcode 16 can be run on macOS 14.5, so we need to
skip the test in that combination.

Reverts llvm/llvm-project#95164

This broke a buildbot:
https://lab.llvm.org/buildbot/#/builders/37/builds/35987

Revert "Support `guarded_by` attribute and related attributes inside C structs and support late parsing them (#94216)"

This reverts commit af0d7128.

Reverting due to likely regression:

https://github.com/llvm/llvm-project/pull/94216#issuecomment-2164013300

The default debug info format for newer versions of Darwin is DWARF 5.


https://developer.apple.com/documentation/xcode-release-notes/xcode-16-release-notes

rdar://110925733

We want the PSTL implementation details to be available regardless of
the Standard mode or whether the experimental PSTL is enabled. This
patch guards the inclusion of the PSTL to the top-level headers that
define the public API in `__numeric` and `__algorithm`.

It can otherwise timeout under some slow configurations.

After 9d0754ad ("[MC] Relax fragments
eagerly") removes the assert of Offset, it is no longer useful to
initialize the member to -1.

Now the symbol value estimate is more precise, which leads to slight
behavior change to layout-interdependency.s.

Summary:
Currently we use `(~0U)` for this definition, however the ~ operator
returns a different sign, meaning that preprocessor checks against this
value will fail. See https://godbolt.org/z/TrjaY1d8q where the
preprocessor thinks that it's not `0xffffffff` while the static
assertion thinks it is. This is because the latter does implicit
conversion but the preprocessor does not. This is now consistent with
other headers.

Reverts llvm/llvm-project#95039

This looks like it caused the ASan bot to fail:
https://lab.llvm.org/buildbot/#/builders/168/builds/20912

Offending line was changed in this PR

Revert "✨ [Sema, Lex, Parse] Preprocessor embed in C and C++ (and Obj-C and Obj-C++ by-proxy)" (#95299)

Reverts llvm/llvm-project#68620

Introduce or expose a memory leak and UB, see llvm/llvm-project#68620

…f weights" #95136

Reverts #95060, and relands #86609, with the unintended code generation
changes addressed.

This patch implements the changes to LLVM IR discussed in
https://discourse.llvm.org/t/rfc-update-branch-weights-metadata-to-allow-tracking-branch-weight-origins/75032

In this patch, we add an optional field to MD_prof meatdata nodes for
branch weights, which can be used to distinguish weights added from
llvm.expect* intrinsics from those added via other methods, e.g. from
profiles or inserted by the compiler.

One of the major motivations, is for use with MisExpect diagnostics,
which need to know if branch_weight metadata originates from an
llvm.expect intrinsic. Without that information, we end up checking
branch weights multiple times in the case if ThinLTO + SampleProfiling,
leading to some inaccuracy in how we report MisExpect related
diagnostics to users.

Since we change the format of MD_prof metadata in a fundamental way, we
need to update code handling branch weights in a number of places.

We also update the lang ref for branch weights to reflect the change.

They were always intended to be in that namespace but I was trying to
keep changes orthogonal.

Implement MVT::getVectorElementType and MVT::getVectorMinNumElements
with table lookup instead of switch.

This speeds up "check-llvm-codegen-amdgpu" by about 7% in my Release
build.

As we allow these nodes to be created pre-legalization, we can't rely on them having a simple VT

Fixes #95278

This patch changes the type of ValueData to
std::vector<InstrProfValueData> so that, in a follow-up patch, we can
teach getValueForSite to return ArrayRef<InstrProfValueData>.

Currently, a typical traversal over the value data looks like:

  uint32_t NV = Func.getNumValueDataForSite(VK, I);
std::unique_ptr<InstrProfValueData[]> VD = Func.getValueForSite(VK, I);
  for (uint32_t V = 0; V < NV; V++)
    Do something with VD[V].Value and/or VD[V].Count;

Note that we need to call getNumValueDataForSite and getValueForSite
separately.  If getValueForSite returns ArrayRef<InstrProfValueData>
in the future, then we'll be able to do something like:

  for (const auto &V : Func.getValueForSite(VK, I))
    Do something with V.Value and/or V.Count;

If ArrayRef<InstrProfValueData> directly points to ValueData, then
getValueForSite won't need to allocate memory with std::make_unique.

Now, switching to std::vector requires us to update several places:

- sortByTargetValues switches to llvm::sort because we don't need to
  worry about sort stability.

- sortByCount retains sort stability because std::list::sort also
  performs stable sort.

- merge builds another array and move it back to ValueData to avoid a
  potential quadratic behavior with std::vector::insert into the
  middle of a vector.

Support `guarded_by` attribute and related attributes inside C structs and support late parsing them (#94216)

This fixes #20777.

Previously the `guarded_by`, `pt_guarded_by`, `acquired_after`, and `acquired_before` attributes were only supported inside C++ classes or top level C/C++ declaration.

This patch allows these attributes to be added to struct members in C. These attributes have also now support experimental late parsing. This is off by default but can be enabled by passing `-fexperimental-late-parse-attributes`. This is useful for referring to a struct member after the annotated member. E.g.

```
struct Example {
  int a_value_defined_before __attribute__ ((guarded_by(a_mutex)));
  struct Mutex *a_mutex;
};
```

Patch by  Pierre d'Herbemont (@pdherbemont)

* Add reductionOperands and reductionAttrs to cuf's KernelOp. 
* Parsing is already working and the tree has the info: here I make the
Bridge emit the updated KernelOp with reduction information added.
* Check |reductionAttrs| = |reductionOperands| in verifier
* Add a test
@clementval @vzakhari

---------

Co-authored-by: Iman Hosseini <imanh@nvidia.com>
Co-authored-by: Valentin Clement (バレンタイン クレメン) <clementval@gmail.com>

Fixes #95271

**DO NOT MERGE** until https://github.com/llvm/llvm-project/pull/89003

getValueForSite computes the total count -- the total number of times
a given value site is visited.  The problem is that, excluding tests,
annotateValueSite is the only place that needs the total count.

This patch moves the total count computation to annotateValueSite.