[flang-commits] [flang] [llvm] [OpenMP][Flang] Fix atomic operations on complex types (PR #165366)

[Krish Gupta via flang-commits]

KrxGu wrote:

@kparzysz Commit message updated and squashed per review. CI all green!!

@kiranchandramohan Re: colocating tests with the fix in OpenMPIRBuilder, plan is:
- Add a gtest in `llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp` that builds atomic write/read on `{double,double}` and asserts the first arg to `__atomic_{load,store}` is 16.
- Add an MLIR translation check in `mlir/test/Target/LLVMIR/openmp-llvm.mlir` for `omp.atomic` lowering to `__atomic_*` with `i64 16`.

I can keep the Flang IR tests as frontend coverage, or drop them if you prefer only the two tests above. Please advise.

@kparzysz Does this test plan align with your expectations? and are we on the same page?


https://github.com/llvm/llvm-project/pull/165366