[llvm] [AMDGPU] Create new option for force flush load counter (PR #124974)

[Austin Kerbow via llvm-commits]

https://github.com/kerbowa approved this pull request.

LGTM.

Only comment is that it's strange that `amdgpu-waitcnt-forcezero` always emits waitcnt after every instruction and this new option waits for some waitcnt to be required before flushing to zero. Maybe we should revamp this option later so that you can select the counter to flush on the cl and also change this behavior. These flags are useful for debugging.

https://github.com/llvm/llvm-project/pull/124974