Add include directory for omp.h at `/opt/IBM/openxlCSDK/include/openmp`.

This allows to configure both the op used for allocation and copy of
memrefs.
It also changes the default behavior because the default allocation in
`BufferizationOptions` creates `memref.alloc` with `alignment = 64`
where we used to create `memref.alloca` without any alignment before.
Fixes
```
// TODO: Use alloc/memcpy callback from BufferizationOptions if called via
// BufferizableOpInterface impl of ToMemrefOp.
```

[clang] Fix a "!CodeSynthesisContexts.empty()" assertion failure when constructing aggregate deduction guides. (#89227)

We were missing to push an record to the instantiation stack in
`DeclareAggregateDeductionGuideForTypeAlias`. This patch fixes that.

When collecting loop scalars, LoopVectorize over-eagerly marks the
induction variable and its update as scalars after vectorization, even
if the induction variable update is a first-order recurrence. Guard the
process with this check, fixing a crash.

Fixes #72969.

Deallocating VGPRs interferes with doing a context save, which is needed for GDB
to report a breakpoint. So, in this sequence:

  s_sendmsg MSG_DEALLOC_VGPRS
  s_endpgm

We now use the debug location of the s_endpgm for the s_sendmsg, so a breakpoint
set in the debugger at the end of a shader will be hit before deallocating VGPRs.

The fix was committed in
https://github.com/llvm/llvm-project/commit/8cee94e989b5bf6fb6455087d48eb6c6e0e23c54.
This adds a test.

These changes were missed in PR #88526.

[DAG] Ensure extract_subvector(insert_subvector(x,y,c1),c2) --> extract_subvector(y,c2-c1) is working on fixed vector types

#87925 failed to ensure we weren't removing the extracted subvector from a scalable vector type

Thanks to @antmox for the headsup.

In preparation for a cleaner addition in a following PR.

This colocates the CPU algorithms closer to the rest of the PSTL
implementation details.

These were added in faf697e4 so things
can flow through non-opaque LLVM ptrs. Those ptrs are gone so there is
no reason for this to be around anymore. LLVM doesn't support f8 types,
they get converted to i8 when lowering to LLVM dialect.

Removing the f8 types makes LLVM::isCompatibleType and
LLVM::isCompatibleFloatingPointType consistent again.

[Clang][Sema] Warn when 'exclude_from_explicit_instantiation' attribute is used on local classes and members thereof (#88777)

A local class and its members declared in a function template are
instantiated alongside the definition of that template. It therefore
does not make sense to apply the `exclude_from_explicit_instantiation`
attribute to such declarations, and this patch adds a warning to
diagnose these cases (in addition to ignoring the attribute).

(The motivation for this patch is to fix a failing test in libc++ for
#84050. In particular, line 199 in `include/__memory/uses_allocator_construction.h`
in libc++ contains the expression `this->__value_` (reduced to 
https://godbolt.org/z/KqEerKWPd) which will be looked up prior
to instantiation once #84050 lands (the lookup context is the 
current instantiation). `_LIBCPP_HIDE_FROM_ABI` includes 
`__attribute__((exclude_from_explicit_instantiation))`, which in the 
reduced example results in `Local` being instantiated with 
`Local::operator A` as its `DeclContext`)

Move the documentation of the ownership-based buffer deallocation pass
to a separate file. Also improve the documentation a bit and insert a
figure that explains the `bufferization.dealloc` op (copied from the
tutorial at the LLVM Dev Summit 2023).

Need to use proper index variable to fix a crash.

This reverts commit e90bc9cf.

This commit improves LLVM dialect's Mem2Reg interfaces to support
promotions of partial loads from larger memory slots. To support this,
the Mem2Reg interface methods are extended with additional data layout
parameters. The data layout is required to determine type sizes to
produce correct conversion sequences.

Note: There will be additional followups that introduce a similar
functionality for stores, and there are plans to support accesses into
the middle of memory slots.

Part 1 of fix for issue
https://github.com/llvm/llvm-project/issues/54624

Split from PR #87623. Clang front end changes to follow.

Use DICompositeType to represent the template alias, using its extraData
field as a tuple of DITemplateParameter to describe the template
parameters.

Added template-alias.ll  - Check DWARF emission.
Modified  frame-types.s  - Check llvm-symbolizer understands the DIE.

This reverts commit 16f18876.

This broke Libcxx Picolib testing at the install step, and builds
for Windows builtins. Revert while we figure out the cause.

This extends on the case from 9c970d5e;
if a section is marked discardable, it won't be mapped into memory at
runtime, so there's no point in creating runtime pseudo relocations for
such sections.

It turns out that the previous name is vaguely misleading.

When operating on a def file like "symbolname == dllname", that is
supposed to make an import library entry, that when the symbol
"symbolname" links against this, it imports the DLL symbol "dllname"
from the referenced DLL. This doesn't need to involve any alias, and it
doesn't need to imply that "dllname" is available on its own as a
separate symbol in the import library at all.

GNU dlltool implements import libraries in the form of "long import
library", where each member is a regular object file with section chunks
that compose the relevant .idata section pieces. There, this kind of
import renaming does not involve any form of aliases, but the right
.idata section just gets a different string than the symbol name.

A bug was found where mops instructions were being generated that
aliased the source and size registers. This is unpredictable behaviour.
This patch uses the earlyclobber constraint on the input source
register so that it doesn't alias with the size register. Also a test is
introduced which checks affected instructions can't violate this
constraint.

This patch enables constant folding for 128 bit floating-point logf
calls. This is achieved by querying if the host system has the logf128()
symbol available with a CMake test. If so, replace the runtime call with
the compile time value returned from logf128.

In the process of collecting instructions to scalarize, LoopVectorize
uses faulty reasoning whereby it also adds instructions that will be
scalar after vectorization. If an instruction satisfies
isScalarAfterVectorization() for the given VF, it should not be appended
to InstsToScalarize. Add this extra guard, fixing a crash.

Fixes #55096.

This patch defines SYS_gettid as __NR_gettid if SYS_gettid is not
available to avoid compile time errors due to SYS_gettid not being
defined. This happens with certain libcs (like bionic) that do not
define SYS_gettid.

Reverts llvm/llvm-project#88965

This caused a test suite failure:
https://lab.llvm.org/buildbot/#/builders/185/builds/6583
NOEXE: test-suite::aarch64-acle-fmv-features.test

```
/home/tcwg-buildbot/worker/clang-aarch64-lld-2stage/test/test-suite/SingleSource/UnitTests/AArch64/acle-fmv-features.c:98:1: error: redefinition of 'check_sha1'
   98 | CHECK(sha1, {
      | ^
/home/tcwg-buildbot/worker/clang-aarch64-lld-2stage/test/test-suite/SingleSource/UnitTests/AArch64/acle-fmv-features.c:36:17: note: expanded from macro 'CHECK'
   36 |     static void check_##X(void) { \
      |                 ^
<scratch space>:150:1: note: expanded from here
  150 | check_sha1
      | ^
```

I presume that the useless features need to be removed from the fmv test
as well.

Although we want to treat the module initializer as a transparent
concept to users, but it shows that people need to understand
the concept to understand how to understand and distribute modules.

So it is better to mention this too.

https://discourse.llvm.org/t/running-lldb-in-a-container/76801/4 found
that the obvious way to use this variable doesn't work, despite what our
docs and examples say.

Perhaps in the past it did but now you need to use ";" as a separator to
make sure the final command line works properly.

For example "-A foo" becomes "-A foo" when python goes to run the runner
script. The script sees this as one command line element, not two. What
you actually want is "-A;foo" which we convert to "-A" "foo" which the
script sees as one option and one value for that option.

The "Script:" printout from dotest is misleading here because it does `"
".join(cmd)` so it looks like it's ok but in fact it's not.

I'm not changing that format though because printing the command as a
Python list is not useful outside of this specific situation.

This PR add `TmaDescriptorBuilder` 
- class simplifies TMA generation.
- Makes the code ready to support various Tma configurations 
   - removes strings and use the enums from `mlir.nvgpu.ENUMs`.
- Example "swizzle = swizzle_128b, l2promo=none, oob=zero,
interleave=none" to enums in `mlir.nvgpu` dialects.
- Enums have string equivalent that are used during the IR writing and
generation (see `TmaDescriptorBuilder::tensormap_descriptor_ty`).
- Improves readability and abstracts out TMA descriptor builders in
reusable component.

---------

Co-authored-by: Manish Gupta <manigupta@google.com>

llvm-project/clang/lib/Serialization/ASTWriter.cpp:5077:59:
error: lambda capture 'SemaRef' is not used [-Werror,-Wunused-lambda-capture]
    auto AddEmittedDeclRefOrZero = [this, &SemaDeclRefs, &SemaRef](Decl *D) {
                                                       ~~~^~~~~~~
1 error generated.

The first argument to the nvvm_shfl_sync_* family
of intrinsics is the thread_mask (aka member_mask). 
This patch renames the corresponding operand in the Op
to reflect the same i.e. `dst` -> `thread_mask`.

While we are there, add summary and description
for this Op.

Signed-off-by: Durgadoss R <durgadossr@nvidia.com>

module purview

For,

```
export module A;
static int impl() { ... }
export int func() { return impl(); }
```

Previously, even with reduced BMI, the function `impl` will be emitted
into the BMI. After the patch, the static entities in module purview
won't get emitted eagerly. Now the static entities may only be emitted
if required.

Note that, this restriction is actually more relaxed than the language
standard required. The language spec said, the program is ill-formed if
any TU-local entities get exposed. However, we can't do this since there
are many static entities in the headers of existing libraries.
Forbidding that will cause many existing program fail immediately.

Another note here is, we can't do this for non-static non-exported
entities, they can be used for other module units within the same
module.

Updated cet test to:
-Check different modules based on different cet options
-Added negative tests also

---------

Signed-off-by: Isha Agarwal <isha.agarwal@intel.com>

It breaks expression evaluation on arm, and the x86 breakage has been
fixed in 6cea7c49.

This reverts commit 915c84b1.

Reverts llvm/llvm-project#88865

There were failing tests in the CI that I didn't notice. Sorry.