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

[Pierre van Houtryve via llvm-commits]

Pierre-vh wrote:

I'm still not sure how this works with out of order counters. This is only useful if you have multiple loads of the same type issued back-to-back and you're trying to find the problematic one, correct?

> This allows profilers to identify which specific memory operation is causing a stall.

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.

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