[Mlir-commits] [flang] [llvm] [mlir] [openmp][mlir] Expand use of atomicrmw for atomic update (PR #215654)

[Ville-Markus Yli-Suutala]

VeeEM wrote:

I was inspired to create this patch from noticing that atomic add to floats is a lot slower with flang than clang.

I added some new tests for affected operations that were not tested before. Two of the old tests that I modified, I thought looked like they are intended to specifically test the cmpxchg loop path, so I changed the operation to keep them as is. The rest of the tests I modified, I updated to check for atomicrmw instead.

https://github.com/llvm/llvm-project/pull/215654