http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50145
--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-08-21 19:37:26 UTC --- Can you please check why? I suppose it's a big/little-endian issue with the testcase, so the expected array contents change? Making the short input array values duplicated in both bytes could fix the test (with adjusting the expected output as well)