http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59025
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Just had a quick look at 3dview.c and indeed, the only swap of arguments I see is due to different SSA_NAME_VERSIONs being used by reassoc1 (that is not a bug) and then during copyprop5 when a stmt is folded using fold_stmt, tree_swap_operands_p is called and that prefers for commutative operands operand with smaller SSA_NAME_VERSION first. So yes, there can be small code generation differences, and there is I'm afraid nothing that can be done about that.
