[clang] [llvm] [AMDGPU] Add an option to disable unsafe uses of atomic xor (PR #69229)

[Yaxun Liu via llvm-commits]

yxsamliu wrote:

> > Will the metadata for unsafe-fp-atomics also be controlled by the pragma that controls the no-fine-grained and no-remote metadata? e.g. something like
> > ```
> > #pragma clang atomics begin no-fine-grained(on) no-remote(on) unsafe-fp(on)
> > ```
> 
> Yes, I would expect something like that for consistency. The most problematic part is the denormal flushing in some address spaces, and secondarily the fixed rounding mode in other cases

we might document that those are not supported for now. if users really need them, introducing more controls to support them

https://github.com/llvm/llvm-project/pull/69229