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

[Jay Foad via llvm-commits]

jayfoad wrote:

> Implemented compiler option `-mamdgpu-expand-waitcnt-profiling` to expand `s_waitcnt` instructions with non-zero counter values into sequences for PC-sampling profiling.

Why would you restrict this to "non-zero counter values"?

Why does this need a new subtarget feature?

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