[flang-commits] [flang] [llvm] [flang][OpenMP] Overhaul implementation of ATOMIC construct (PR #137852)

[Krzysztof Parzyszek via flang-commits]

kparzysz wrote:

> So the only thing left from my perspective is the semantics error for 0481_0038. Do you agree this error is a false positive?

I looked through the compile log and there are indeed compilation errors in 04xx, but 0481_0038 is the only regression from main.

The problematic code is `yy=max(1,1,1 ,yy+1)`, specifically the `yy+1` in the argument list.  The OpenMP spec requires the atomic variable itself to be an argument to an intrinsic function in an update operation, but here it's a proper subexpression of an argument.  Given that, this is a legitimate diagnostic message  This requirement exists in all specs going back to 5.0, so I think the test should be updated.

https://github.com/llvm/llvm-project/pull/137852