http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59985
--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> --- It seems the reason for the different chosen alternative for the last anddi3_insn is reload_sum. overall, losers and reload_nregs are the same for nalt 4 and 5, 4 is currently best alternative in both cases, but reload_sum is in one case 0 and in one case 1. bb_reload_num is 13 in one case and 14 in the other case, while last_reload is 14 in both cases.
