[clang] [CIR] Atomic fetch-and-update operations on pointers (PR #195537)

[Erich Keane via cfe-commits]

erichkeane wrote:

> Ooof... you and I kinda 'raced' here: #198871
> 
> I hadn't realized this existed, and fixed it a different way, and our results are slightly different, so I think one of us needs to revert. I have a bit of a preference for MY solution (As I think it handles 'more' cases, though fails to remove the NYI for the 1 set, plus is a bit less 'special casing') , but am obviously biased here :)
> 
> @andykaylor : Can you decide which should be reverted? I can do the work to make sure that the test cases will all pass (currently failing CI) and revert whichever solution you don't prefer.

I'm still looking at this, and I think my opinion has wavered, and I think this is more 'right' in a few places.  I have a patch I'm going to submit in the very near future that should fix up the build issues by updating the test + removing much of what my patch did.  Sorry @Lancern for the issues, I didn't realize this existed :/

https://github.com/llvm/llvm-project/pull/195537