https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71491
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> --- Created attachment 39875 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=39875&action=edit patch In both cases we fail because unaligned loads (or stores) are not supported. Indeed while the testcases make sure the arrays are aligned the variable striding makes it not so in all cases (and GCC fails to analyze it so for some cases where it is). I suppose XFAILing / providing alternat scan-dump for [!] vect_hw_misalign will fix this. Can you test the attached?
