https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91683
--- Comment #8 from Jim Wilson <wilson at gcc dot gnu.org> --- (In reply to Richard Biener from comment #3) > Hmm, but shouldn't we instead fix combine to record not the reg rtx but the > regno for UNDO_MODE? Because nothing prevents this from happening elsewhere > (even in combine itself?). Changing do_SUBST_MODE to save the regno instead of the address of rhe reg rtx is a possible solution, but can work only if there is no other place in combine that might take the address of a reg rtx. I can't be sure of that unless I do more analysis, which I don't plan to do as I don't think this is the right solution. Also, I think I will get better code if I fix the splitters to use a clobber. As was mentioned in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91635#c19 , calling gen_reg_rtx means no REG_DEAD note which limits what combine can do, whereas using a clobber means getting an existing reg that likely already has a REG_DEAD note which helps combine.
