https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109093
--- Comment #33 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Anyway, the problem is as described earlier, in r14-4971 we have 2 pushes after movq %rsp, %rbp, in r14-4972 and onwards just one, so leaq -80(%rsp), %rsi is again not sufficiently aligned for the vmovdqa %xmm2, (%rsi) store (etc.).
