This lets users use TensorDialect without depending on all dialects
transitively. 513cdb82 forces that
otherwise.

--load-bitcode-into-experimental-debuginfo-iterators

      false: Convert to the old debug mode after reading.
      true: Upgrade to the new debug info format (*).
      unset: Same as false (for now).

(*) As of this patch it actually just means "don't convert to either
mode after loading". Auto-upgrading will be implemented in an upcoming
patch.

With this flag we can incrementally add support for RemoveDIs by
overriding the "unset" behaviour in individual tools. The flag can be
removed once all tools support the new debug info mode.

We add a common class `SingleFusion` that accepts a single instruction
pair to simplify fusion definitions.

Pull Request: https://github.com/llvm/llvm-project/pull/85750

`llvm::MaybeAlign` does this, for example.

It's not an option to simply ignore these derived classes because they
get cast
back to the optional classes (for example, simply when calling the
optional
member functions), and our transfer functions will then run on those
optional
classes and therefore require them to be properly initialized.

This patch moves some code in PFT to MLIR OpenMP lowering to the
`ClauseProcessor` class. This is so that some behavior that is related
to certain clauses stays within the `ClauseProcessor` and it's not the
caller the one responsible for always doing this when the clause is
present.

Simultaneously improved diagnostic testing for the `PseudoLI_W`
instruction.

This caused following warnings in an LLDB build:
```
[237/1072] Building CXX object tools/l...lusLanguage.dir/LibCxxSliceArray.cpp.o
/Volumes/Data/llvm-project/lldb/source/Plugins/Language/CPlusPlus/LibCxxSliceArray.cpp:38:53: warning: format specifies type 'unsigned long long' but the argument has type 'size_t' (aka 'unsigned long') [-Wformat]
   38 |   stream.Printf("stride=%" PRIu64 " size=%" PRIu64, stride, size);
      |                         ~~~~~~~~~                   ^~~~~~
/Volumes/Data/llvm-project/lldb/source/Plugins/Language/CPlusPlus/LibCxxSliceArray.cpp:38:61: warning: format specifies type 'unsigned long long' but the argument has type 'size_t' (aka 'unsigned long') [-Wformat]
   38 |   stream.Printf("stride=%" PRIu64 " size=%" PRIu64, stride, size);
      |                                          ~~~~~~~~~          ^~~~
2 warnings generated.
```

This patch simply changes the format specifiers to use the `%zu` for
`size_t`s.

This code was assuming that the LHS would always be one of
GlobalVariable, BlockAddress or ConstantExpr. However, it can
also be a special constant like dso_local_equivalent or no_cfi.
Make sure this is handled gracefully.

Followup to #85188

Update PromoteAllocaToVector so it considers the whole function before promoting allocas.
Allocas are scored & sorted so the highest value ones are seen first. The budget is now per function instead of per alloca.

Passed internal performance testing.

Corrects the `AMDHSA_BITS_SET` macro.

In this patch some uses of `llvm::SmallVector` in Flang's lowering to
MLIR are replaced by other types (i.e. `llvm::ArrayRef` and
`llvm::SmallVectorImpl`) which are intended for these uses. This
generally prevents relying on always passing small vectors with a
particular number of elements in the stack.

Polymorphic entity lowering status is good. The main remaining TODO is
to allow lowering of vector subscripted polymorphic entity, but this
does not deserve blocking all application using polymorphism.

Remove experimental option and enable lowering of polymorphic entity by
default.

This reverts commit d9c31ee9.

This patch adds a couple of new directive sets for composite constructs,
completes some of the existing ones with missing values, refactors all*
sets to always build on the corresponding top* set and reorders sets and
directives alphabetically.

No functional change intended.

The current lowering did not handle sequence associated argument passed
by descriptor. This case is special because sequence association implies
that the actual and dummy argument need to to agree in rank and shape.
Usually, arguments that can be sequence associated are passed by raw
address, and the shape mistmatch is transparent. But there are three
cases of explicit and assumed-size arrays passed by descriptors:
 - polymorphic arguments
 - BIND(C) assumed-length arguments (F'2023 18.3.7 (5)).
 - length parametrized derived types (TBD)

The callee side is expecting a descriptor containing the dummy rank and
shape. This was not the case. This patch fix that by evaluating the
dummy shape on the caller side using the interface (that has to be
available when arguments are passed by descriptors).

We use the VSCBIQ/VSBIQ/VSBCBIQ family of instructions to implement
USUBO/USUBO_CARRY for the i128 data type.  However, these instructions
use an inverted sense of the borrow indication flag (a value of 1
indicates *no* borrow, while a value of 0 indicated borrow).  This
does not match the semantics of the boolean "overflow" flag of the
USUBO/USUBO_CARRY ISD nodes.

Fix this by generating code to explicitly invert the flag.  These
cancel out of the result of USUBO feeds into an USUBO_CARRY.

To avoid unnecessary zero-extend operations, also improve the
DAGCombine handling of ZERO_EXTEND to optimize (zext (xor (trunc)))
sequences where appropriate.

Fixes: https://github.com/llvm/llvm-project/issues/83268

Followup to #85188.

Intend to fix the `Test documentation build `, degraded here
https://github.com/llvm/llvm-project/pull/83149 .

[AArch64] Add support for Cortex-A520AE and Cortex-A720AE CPUs

Cortex-A520AE and Cortex-A720AE are Armv9.2 AArch64 CPUs.

Technical Reference Manual for Cortex-A520AE:
   https://developer.arm.com/documentation/107726/latest/

Technical Reference Manual for Cortex-A720AE:
   https://developer.arm.com/documentation/102828/latest/

Currently, `simplifyMul()` asserts that either `lhs` or `rhs` is
symbolic or constant. This method is called by the overloaded `*`
operator for `AffineExpr`s which leads to a crash when building a
multiplication expression where neither operand is symbolic or constant.
This patch returns a `nullptr` from `simplifyMul()` to signal that the
expression could not be simplified instead.

Fix https://github.com/llvm/llvm-project/issues/75770

This is no longer needed with opaque pointers.

This doesn't change functionality, but lets us avoid attaching all the
interfaces after 513cdb82 turned casting
without loading into an error.

Fixes ompx_dump_mapping_tables test by only using one device after
breaking built bots

If we had a comparison to a literal nan with a false predicate,
we were incorrectly treating it as an unordered compare. This was
correct for fcmp true, but not fcmp false. I noticed this in the
review for e44d3b3e but misdiagnosed
the reason. Also change the test for the fcmp true case to be more
useful, but it wasn't wrong previously.

This patch enables more optimization after canonicalizing `fmul X, 0.0`
into a copysign.
I decide to implement this fold in InstCombine because
`computeKnownFPClass` may be expensive.

Alive2: https://alive2.llvm.org/ce/z/ASM8tQ

Fixes: #84749

To make structure clear.

Reviewers: dtcxzyw, arsenm

Reviewed By: arsenm

Pull Request: https://github.com/llvm/llvm-project/pull/85748

This is a relatively rare case, but

- It's still nice to get this right,
- We can remove the special case for this in
`VisitCXXOperatorCallExpr()` (that
  simply bails out), and
- With this in place, I can avoid having to add a similar special case
in an
  upcoming patch.

This expresses better what the class actually does, and it reduces the
number of
`Context`s that we have in the codebase.

A deprecated alias `ControlFlowContext` is available from the old
header.

This reverts the changes from 91a38462
for Windows targets. The changes in that commit don't work as expected
for Windows targets (those parts of llvm_add_library don't quite behave
the same for Windows), while the previous status quo (producing a
library named "libLLVM-<major>.dll") is the defacto standard way of
doing versioned library names there, contrary to on Unix.

After that commit, the library always ended up named "libLLVM.dll",
executables linking against it would reference "libLLVM.dll", and
"libLLVM-<major>.dll" was provided as a symlink.

Thus revert this bit back to as it were, so that executables actually
link against a versioned libLLVM, and no separate symlink is needed.

The only thing that might be improved compared to the status quo as it
was before these changes, is that the import library is named
"lib/libLLVM-<major>.dll.a", while the common style would be to name it
plainly "lib/libLLVM.dll.a" (even while it produces references to
"libLLVM-<major>.dll", but none of these had that effect for Windows
targets.

(As a side note, the llvm-shlib library can be built for MinGW, but not
currently in MSVC configurations.)

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).

This commit fixes the remaining memory leaks in the MLIR test suite.
`check-mlir` now passes when built with ASAN.