[llvm] [RFC] Memory Model Relaxation Annotations (PR #78569)

[Pierre van Houtryve via llvm-commits]

Pierre-vh wrote:

Note I also fixed AtomicExpand, so now there is (seemingly) no more MMRAs dropped during codegen!
Functionality wise this is pretty much done. Testing is done too I believe.
The biggest ? left is how to handle GVNHoist/SimplifyCFG & other passes that do a lot of moving/combining of instructions.

I believe the best thing to do to preserve MMRA is to require a strict equality, instead of combining. At least for common operations like hoisting/simplifyCFG. However this may be a heavy price to pay because those are core optimizations and inhibiting them too much could eventually be problematic.

https://github.com/llvm/llvm-project/pull/78569