https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90676
Martin Sebor <msebor at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|ambiguous GIMPLE after |[9 Regression] ambiguous |store merging |GIMPLE after store merging --- Comment #2 from Martin Sebor <msebor at gcc dot gnu.org> --- The GIMPLE needs to show the type of the object being stored regardless of endianity, for example like GCC 8 does: MEM[(char[8] *)&a] = 1; Otherwise, there's no way to tell from MEM[(char *)&a] = 1; that more than 1 byte is being stored.