https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68990
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- BTW, with -mno-stv -O3 the testcase passes, and while the *.stv pass actually doesn't do anything at all on this testcase, that switch already changes expansion in a couple of places so no wonder that the RA makes different decisions. And, I haven't managed to create a testcase that would abort vs. return 0 from main for good vs. bad, even other fail defining a stdarg function like printf didn't help. Do, unless somebody comes up with one, guess we'll need a dg-output directive and try to match what is printed (without the newline in there).