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

[Pankaj Dwivedi via llvm-commits]

PankajDwivedi-25 wrote:

> I'm still not sure how this works with out of order counters.

I see the point, you mean the event which share the same counter might finish out-of-order right? this will cause issue because I haven't made any check here. 

Is it problematic or it is still useful for profiling purposes?

One quick fix that arise in my mind is skip expanding out of order events. exploring if we can also check if they are really different events rather than just returning true if more than one bit is set. @Pierre-vh 

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