Thanks both. v2 will be a single store-merging patch: no new hook, and STRICT_ALIGNMENT and TARGET_SLOW_UNALIGNED_ACCESS are left alone. The pass asks about its own transform instead: on a strict-alignment target it widens a store, or the load feeding it, below its alignment only where movmisalign costs no more than the aligned accesses it replaces. arm keeps its 416-byte firmware saving with no backend change; alpha with BWX is unchanged, as its byte stores are cheaper, and alpha without BWX merges and gets smaller. I'm happy to help with PR107626 separately.
Dominic
