topperc wrote: > LGTM. > > On another note, have we looked at using the MacroFusionDAGMutation for PostMachineSched? It's handled in `llvm::createGenericSchedPostRA` already https://github.com/llvm/llvm-project/pull/117527