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

[Pierre van Houtryve via llvm-commits]

Pierre-vh wrote:

> Thanks for the comments @arsenm @yxsamliu @b-sumner.
> 
> By approaching a similar solution, do you mean MMRAs (#78569) ?
> 
> If so, should I rebase/adapt my patch to the MMRA PR? Or will this PR be redundant and needs closing?
> 
> @yxsamliu These concise names look good to me.

In this case, MMRAs would only help in the sense that you won't need any new attributes and can just add an MMRA such as `atomic-lowering:fine-grained`. It's not really what MMRAs were made for (because this attribute doesn't affect semantics, just lowering style I think?), but all of the boilerplate would be there to preserve this annotation until the backend, and as long as you don't add any other `atomic-lowering:<X>` flag, you shouldn't have issues with compatibility relations - if you plan to add other flags like that then it may be an issue because you could end up with two operations being considered incompatible (= can reorder them) when they're still technically compatible

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