https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123853
--- Comment #17 from Richard Sandiford <rsandifo at gcc dot gnu.org> --- FWIW, I agree with Drea in comment 9 that the transformation appears to be valid in RTL terms, since the substitution happens in an address. And I agree that it's up to the m68k port to reject this pattern if m68k pre-inc moves don't have the behaviour that the RTL docs describe. rtl-ssa/changes.cc:recog_level2 already looks for uses of RTX_AUTOINC registers outside of an address.
