https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127865
--- Comment #2 from Jeffrey A. Law <law at gcc dot gnu.org> --- So just a quick update. I've go a change to the AOP expander that fixes this problem. There's a generic question I've got out to Roger about one aspect of how the AOP code is supposed to work -- it's showing up in REG_EQUAL notes and not surprisingly triggering gcc_assertions in simplify-rtx. I think that's unintentional (and assuming so pretty trivial to fix). With the expander change and avoiding AOP in REG_EQUAL notes, this problem will be fixed. However, I won't be at all surprised if we find cases where it regresses the other way. In particular things like setting a single bit that won't fit in a simm12. That's easier to detect as a bset via IOR than PLUS. Just something to keep in mind.
