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

[via cfe-commits]

b-sumner wrote:

We're aware of the concerns regarding atomics handling.   We're trying to find the approach which is least distasteful to all parties.  I expect we will address the concern raised here, but not necessarily in the same way.

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