https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90883
--- Comment #8 from Jeffrey A. Law <law at redhat dot com> --- Oh, yea, I kept looking at this from a DSE lens in which case it's the earlier store that is partially dead. But if we're storing the same value, then the latter store is totally dead and removing the latter store is a better choice. This might actually be fairly easy to implement, let me poke around a bit. WRT the padding byte and the unaligned issues. We try to keep the head of an area aligned, but it's much less important for the tail. And I even poked at things in the debugger and having DSE ignore that padding byte doesn't really help. Really the way to go is to realize the second store is redundant because it's storing the exact same values as the earlier store.