Existing reduction detection algorithm does two types of memory checks
before marking a load store pair as reduction.
Second check is to verify there is no other memory access in ScopStmt
overlapping with the memory of load and store that forms the reduction.
Existing check misses cases where there could be probable overlap such
as
	A[V] += A[P];
In the above case there is chance of overlap between A[V] and A[P] which
is missed.
This commit addresses this by removing the parameter from space before
checking for compatible space.

Part 1 of this patch :
[75297](https://github.com/llvm/llvm-project/pull/75297)

- AMDGPU/NFC: Purge SOPK_ZEXT from TSFlags
  - Moved to helper function in SIInstInfo
- AMDGPU/NFC: Purge VOPAsmPrefer32Bit from TSFlags
  - This flag did not make sense / remnants of something else I think

PR #71709 broke the Linux PIE build with `undefined symbol: alloca`
errors. With the newly included `clang/Basic/Builtins.h` in that PR, it
surfaces an issue with a combination of two previous patches.

26670dcb added `#undef alloca` so clang
builtins handling of alloca would work under MSVC (unsure if this is
still necessary).

194b6a3b added code that calls `alloca`
to workaround a Linux kernel < 4.1 bug. Given that Linux 4.1 was EOL in
2018, it should be ok to remove this workaround.

Apply clang-tidy fixes for performance-unnecessary-value-param in BufferizableOpInterfaceImpl.cpp (NFC)

Add support for attribute nvvm.grid_constant on LLVM function arguments.
The attribute can be attached only to arguments of type llvm.ptr that
have llvm.byval attribute.
Generate LLVM metadata for functions with nvvm.grid_constant arguments.
The metadata node is a list of integers, where each integer n denotes
that the nth parameter has the
grid_constant annotation (numbering from 1). The generated metadata node
will be handled by NVVM compiler. See
https://docs.nvidia.com/cuda/nvvm-ir-spec/index.html#supported-properties
for documentation on grid_constant property.

This patch also adds convertParameterAttr to
LLVMTranslationDialectInterface for supporting the translation of
derived dialect attributes on function parameters 

The replacement should've had BFE() as the arguments for the comparison,
not the source register.

While at that, tighten the patterns a bit, and expand them to cover
variants with immediate arguments. Also change the default lowering of
bfe() to use unsigned variant, so the value of the upper bits is
predictable.

This refactors the fast-math handling in the clang driver, moving the
settings into a lambda that is shared by the -ffp-model=fast and
-ffast-math code. Previously the -ffp-model=fast handler changed the
local option variable and fell through to the -ffast-math handler.

This refactoring is intended to prepare the way for decoupling the
-ffp-model=fast settings from the -ffast-math settings and possibly
introduce a less aggressive fp-model.

Switch the primary implementation of EPC lookupSymbols to be async,
keeping a synchronous wrapper for compatibility. Use the new async
implementation inside EPCDynamicLibrarySearchGenerator to work
working towards a fully async search generator.

Provide an asynchronous lookup API for EPCGenericDylibManager and adopt
that from the SimpleRemoteEPC. This enables an end-to-end async
EPCDynamicLibrarySearchGenerator. Note: currently we keep the current
per-dlhandle lookup model, but a future improvement could do a single
async call for a given lookup operation.

This refactored pattern rewrite is intended to be reused by any
`LinalgOp`'s canonicalization pattern for removing identity ops.
Additionally, this canonicalization has been applied to `BroadCastOp`.

Turns out this arch is needed by the ABI impls

rdar://121963634



Co-authored-by: Mariusz Borsa <m_borsa@apple.com>

As noted in issues #68594 and #73935, `JITLink/RISCV/ELF_ehframe.s`
fails with libstdc++'s expensive checks because `getRISCVPCRelHi20`
calls `std::equal_range` on the edges which may not be ordered by their
offset. Instead let `ELFJITLinker_riscv` build a hashmap of all edges
with type `R_RISCV_PCREL_HI20` that can be looked up in constant time.

Closes #73935

Row and column arguments for matrix_transpose indicate the shape of the
operand. When hoisting the transpose to the result of the add, the add
operates on the original operand's shape, and so does the hoisted
transpose.

This patch also adds an assert that the shape for the original add and
the transpose match, as well as the shape of the new add matches the
cached shape for it.

The assert could potentially be moved to
updateShapeAndReplaceAllUsesWith.

Covers CWG issues
[124](https://cplusplus.github.io/CWG/issues/124.html)
[185](https://cplusplus.github.io/CWG/issues/185.html),
[193](https://cplusplus.github.io/CWG/issues/193.html),
[199](https://cplusplus.github.io/CWG/issues/199.html).

I also looked at [190](https://cplusplus.github.io/CWG/issues/190.html),
but concluded that we should try to test it via C++20
`std::is_layout_compatible` first.

I tried to group tests under `dr1xx-codegen.cpp`, but found out that
CodeGen can arbitrarily reorder function definitions in LLVM module. In
particular, interleaving between regular function definitions and
destructor definitions present in the source might not be preserved,
which messes up FileCheck directives. `CHECK-DAG` can help with that,
but its interaction with `CHECK-LABEL` (lack of thereof) would require
me to relax tests too much.

The mask for the reshuffling/resizing might be calculated incorrectly,
fixed.

Added basic support for strided loads support in SLP vectorizer.
Supports constant strides only. If the strided load must be
reversed, applies -stride to avoid extra reverse shuffle.

Reviewers: preames, lukel97

Reviewed By: preames

Pull Request: https://github.com/llvm/llvm-project/pull/80310

In the beginning, Clang only emitted atomic IR for operations it knew
the
underlying microarch had instructions for, meaning it required
significant
knowledge of the target. Later, the backend acquired the ability to
lower
IR to libcalls. To avoid duplicating logic and improve logic locality,
we'd like to move as much as possible to the backend.

There are many ways to describe this change. For example, this change
reduces the variables Clang uses to decide whether to emit libcalls or
IR, down to only the atomic's size.

The DWARF 5 debug_str_offsets section starts with a header, which must
be skipped in order to access the underlying `strp`s.

However, the verifier supports some pre-standardization version of this
section (with the same section name), which does not have a header. In
this case, the offsets start on the first byte of the section. More in
[1] and [2] about this legacy section.

How does The DWARF verifier figure out which version to use? It manually
reads the **first** header in debug_info and uses that. This is wrong
when multiple debug_str_offset sections have been linked together, in
particular it is wrong in the following two cases:

1. A standard DWARF 4 object file (i.e. no debug_str_offsets) linked
with a standard DWARF 5 object file.
2. A non-standard DWARF 4 object file (i.e. containing the header-less
debug_str_offsets section) linked with a standard DWARF 5 object file.

Based on discussions in https://github.com/llvm/llvm-project/pull/81210,
the legacy version is only possible with dwo files, and dwo files cannot
mix the legacy version with the dwarf 5 version. As such, we change the
verifier to only check the debug_info header in the case of dwo files.
If it sees a dwarf 4 version, it handles it the legacy way.

Note: the modified test was technically testing an unsupported
combination of dwarf version + non-dwo sections. To see why, simply note
that the test contained no `debug_info.dwo` sections, so the call to
DWARFObject::forEachInfoDWOSections was doing nothing. We were finding
the error through the "standard version", which shouldn't happen.

[1]: https://gcc.gnu.org/wiki/DebugFission 
[2]: https://gcc.gnu.org/wiki/DebugFissionDWP

This patch converts `Sema::TemplateDeductionResult` into a scoped enum
in namespace scope, making it eligible for forward declaring. This is
useful in certain contexts, such as `preferred_type` annotations on
bit-fields.

This commit brings support for debug_names in DWARFYAML. It parses YAML
and generates emits a DWARF5 Accelerator table with the following
limitations:

1. All forms must have a fixed length (zero length is also ok).
2. Hard-coded support for DWARF 5 and DWARF32.
3. The generated table does not contain a hash index

All of these limitations can be lifted in the future, but for now this
is good enough to enable testing.

This patch reworks the way that parallel reduction operations function
to better match the expected semantics from the OpenMP specification.
Previously specific omp.reduction operations were used inside the
region, meaning that the reduction only applied when the correct
operation was used, whereas the specification states that any change to
the variable inside the region should be taken into account for the
reduction.

The new semantics create a private reduction variable as a block
argument which should be used normally for all operations on that
variable in the region; this private variable is then combined with the
others into the shared variable. This way no special omp.reduction
operations are needed inside the region.

This patch only makes the change for the `parallel` operation, the
change for the `wsloop` operation will be in a separate patch.

---------

Co-authored-by: Kiran Chandramohan <kiran.chandramohan@arm.com>

The bug mentioned in the comment has been committed and did change the
cfi_offset.

When a branch target is too far away we need to emit an indirect branch.
We scavenge a register for this since we don't know we need this until
after register allocation.

Jumps using X1 and X5 as the source are hints to the hardware to pop the
return-address stack. We should avoiding using them for jumps that
aren't a return or tail call.

Follow-up for #78374

These hardcoded attribute name are a leftover from the upstreaming
period when there was no way to get the attribute name without an
instance of the operation. It is since possible to do without them and
they should be removed to avoid duplication.

This PR cleanup the fir.dt_entry and fir.dispatch ops of these hardcoded
attribute name and use their generated getters. Some other PRs will
follow to cleanup other operations.

These hardcoded attribute name are a leftover from the upstreaming
period when there was no way to get the attribute name without an
instance of the operation. It is since possible to do without them and
they should be removed to avoid duplication.

This PR cleanup the fir.global op of these hardcoded attribute name and
use their generated getters. Some other PRs will follow to cleanup other
operations.

This reverts commit 0940f908 to fix
issues reported in https://github.com/llvm/llvm-project/pull/80310.