https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71734
--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Created attachment 38930 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=38930&action=edit gcc7-pr71734.patch The bug is of course in using (uselessly) an x86_64/i?86 specific intrinsic headers in a generic test on all architectures. There is no need to use that header though, it fails the same way with __builtin_posix_memalign/__builtin_free, and as it is a compile time only test, we don't care if the library implements it or not.