[llvm] [AMDGPU] Implement Waitcnt Expansion for Profiling (PR #169345)

[Jay Foad via llvm-commits]

jayfoad wrote:

> Memory operations to different address can complete in any order, even if the waitcnt is decremented in a fixed order. I'm not sure of how useful this is in practice.

"the waitcnt is decremented in a fixed order" is all you need: that allows you to associate each s_waitcnt instruction with a corresponding load instruction (at least in straight line code). I can see how that would be useful for profiling.

https://github.com/llvm/llvm-project/pull/169345