This selects non-PreISelGenericOpcode as-is.

Co-authored-by: Origami404 <Origami404@foxmail.com>

This adds initial codegen support for BPF backend.

Only implemented ir-translator for "RET" (but not support isel).

Depends on: #74998

The predicate should match the operand types.

Add missing rm postfix to show these are load instructions

This must be an integer. This check is present in LLParser but not
the BitcodeReader.

This fixes #75010.

We can tidy up the interfaces a bit once all intrinsics are supported, as we
will have a more informed view then.

Fixes https://github.com/llvm/llvm-project/issues/74453

.

The `gepToByteOffset` was implicitly casting an signed integer to an
unsigned integer even though negative dimensions are valid for
`llvm.getelementptr`.

---------

Co-authored-by: Tobias Gysi <tobias.gysi@nextsilicon.com>

This was moved to VectorToArmSME in #74063, so this is no longer needed.

VectorToArmSME uses a greedy rewriter, so a similar legality rule is not
needed there.

See:
https://github.com/llvm/llvm-project/blob/bbb8a0df7367068e1cf2fc54edd376beb976b430/mlir/lib/Conversion/VectorToArmSME/VectorToArmSMEPass.cpp#L35

The below changes were made:
- test the vector and scalar names, the ISA, and the presence or absence of a
  mask in all tests that is relevant
- test the number of the parameters, the order/types and for masks if present
- replaced methods like `sin` to `foo` to make it more clear that these are not
  existing functions but they are rather tests.
- using mostly `i32` for parameters where it is not relevant, except when the VF
  of elements in explicitly checked, and `ptr` for references.
  Also using `void` for return types.
- all `VFABIParserTest` tests are now listed contiguously in the source.
- Removed duplicate ISA tests
- Added an extra test to clearly show that the mangled name becomes the
  VectorName, when no VectorName is specified.
- Use `VFInfo` for `isMasked`
- Minor code refactoring, cleanup, and improved comments

Without this patch, MLIR crashes with
```
Assertion failed: (getNumDims() == map.getNumResults() && "Number of results mismatch"), function compose, file AffineMap.cpp, line 537.
```
during parsing.

There was duplicated (and complex) code querying whether tags were
type-like tags (i.e. class or struct); this has been factored out into a
helper function.

There was also a comment about not comparing identical DIEs without ever
performing that check; this comment has been removed. It was likely a
result of copy paste from another function in this same file which
actually does that check.

Fixes SWDEV-421067

UserDefinedLiteral is also a sub class of CallExpr but we forgot to
initialize it in the same way as other sub classes of CallExpr.

Types of AMDGPU address space were defined not only in Clang-specific class
but also in LLVM header.

If we unify the AMD GPU address space enumeration, then we can reuse it in
Clang, Flang and LLVM.

move constructors should be marked noexcept

If the loop bound is not initialized, the analysis crashed, as it only checked for nullity. Also checking for initialization fixes the issue.

Signed-off-by: Victor Perez <victor.perez@codeplay.com>
Co-authored-by: Tsang, Whitney <whitney.tsang@intel.com>

This is used for eliminate uses of "CurDAG", which is SelectionDAG-spec,
and not compatible with GIsel algorithms.

(NFC)

This patch tries to reduce the size of the BMIs by packing more bits
into an unsigned integer.

This patch was reverted due to buildbot failure report. But it should be
irrevelent after I took a double look. So I tried to recommit this NFC
change again.

Added the following functions for manipulating operand bundles, as well as
building ``call`` and ``invoke`` instructions that use operand bundles:

  * LLVMBuildCallWithOperandBundles
  * LLVMBuildInvokeWithOperandBundles
  * LLVMCreateOperandBundle
  * LLVMDisposeOperandBundle
  * LLVMGetNumOperandBundles
  * LLVMGetOperandBundleAtIndex
  * LLVMGetNumOperandBundleArgs
  * LLVMGetOperandBundleArgAtIndex
  * LLVMGetOperandBundleTag

Fixes #71873.

As part of #73798 there was some discussion about using the format
helper to run from a git-hook. That was not possible for a number of
reasons, but with these changes it can now be installed as a hook and
then run on the local cache in git instead of a diff between revisions.

This also checks for two environment variables DARKER_FORMAT_PATH and
CLANG_FORMAT_PATH where you can specify the path to the program you want
to use.

Support loops without static boundaries. Since the number of iteration
is not known we need to predicate prologue and epilogue in case the
number of iterations is smaller than the number of stages.

This patch includes work from @chengjunlu

This reverts commit 10951050.
This should be part of 8c334627 to revert
9406ea3f completely.

This reverts commit 9406ea3f.

There are build bots complaining this. Revert it first to try to keep
the bots green.

Forgot to rename `winehprepare` for legacy pass when port this pass to
new passmanager.

The member functions of ScheduleDAGMI are called back from
PostMachineScheduler::runOnMachineFunction, instead of
MachineScheduler::runOnMachineFunction.

Replaced explicit loops with find + erase.

llvm-project/clang/lib/Serialization/ASTWriterDecl.cpp:2342:12:
 error: unused variable 'ExprDependenceBits' [-Werror,-Wunused-variable]
 2342 |   unsigned ExprDependenceBits = llvm::BitWidth<ExprDependence>;
      |            ^~~~~~~~~~~~~~~~~~
1 error generated.

If a local stub is out-of-range, at LongJmp we will try to find another
local stub first. However, The original implementation do not work as
expected and it leads to an infinite loop between replaceTargetWithStub
and fixBranches.

After this patch, we first convert the target of BB back to the target
of the local stub, and then look up for other valid local stubs and so
on.

This is similar to single and double-precision floating-point
instructions.