Old intrinsics were forcing these, but they have now all been removed. This fixes large i8 vector operations generally being broken. llvm-svn: 258788