https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65146
--- Comment #6 from Peter Cordes <peter at cordes dot ca> --- My test-case on godbolt: https://godbolt.org/g/MmLycw. gcc8 snapshot still only has 4B alignment Fun fact: clang4.0 -m32 inlines lock cmpxchg8b for 8-byte atomic load/store. This is ironic, because it *does* align _Atomic 64-bit objects to 8 bytes so it could safely use SSE loads/stores. It would work correctly if called from gcc-compiled code that passed it a misaligned atomic_llong *. But since gcc and clang don't agree on i386 SysV struct layout for _Atomic 64-bit members, so clang should really just start using movq for 64-bit atomic objects in 32-bit mode.
