[fweimer]

It makes more sense now that Intel and AMD retconned naturally-aligned 128-bit atomic loads into the ISA: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104688 (AMD's confirmation is in comment 10.)