https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121322
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Reduced testcase: unsigned long long foo (unsigned long long *p) { unsigned long long a = *p; unsigned long long b = __builtin_bswap64 (a); return ((b << 32) | ((b >> 8) & 0xff000000ULL) | ((b >> 24) & 0xff0000ULL) | ((b >> 40) & 0xff00ULL)); } tmp_n is 0x403020105060700, the zero LSB byte is ignored and because the next one is 7, (7 + 1) % 8 is 0. We've already tried that and that didn't work, so we should just punt.