GLib's macros are implemented with inline assembly, and probably a lot
faster than our naive bit shifting implementation.