http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59461
--- Comment #3 from Eric Botcazou <ebotcazou at gcc dot gnu.org> --- > I wonder if x86_64 is also affected as it has implicitely zero/sign-extending > loads as well. Not for this testcase at least, where the code is (and has always been) optimal: ee_isdigit2: .LFB0: .cfi_startproc movl %edi, %edi movzbl zeb_test_array(%rdi), %eax subl $48, %eax cmpb $9, %al setbe %al ret .cfi_endproc because the x86-64 can perform the addition in QImode directly.