[llvm] [AMDGPU] Register allocation anti-hints to reduce MFMA hazard NOPs (PR #156943)

[via llvm-commits]

================

----------------
LU-JOHN wrote:

Do all the `CurrentCycle` maintenance here. Set:
`unsigned PrevInstrNumWaitStates = 0;`
outside the loop. Then just do:
```
CurrentCycle+=PrevInstrNumWaitStates;
PrevInstructionNumWaitStates = SIInstrInfo::getNumWaitStates(MI);
```
here.



https://github.com/llvm/llvm-project/pull/156943