https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69614
--- Comment #8 from Jeffrey A. Law <law at redhat dot com> --- I'd have to look at the current DCE implementation to if it's still relevant -- in the past we used to emit clobbers like this to tell DCE that the clobbered value is totally written when the writes were split up into upper/lower halves. I don't see anything inherently wrong with this peephole, changing the condition like you did is probably just making the underlying real problem.