https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78102
--- Comment #3 from Marc Glisse <glisse at gcc dot gnu.org> --- (In reply to Richard Biener from comment #2) > This means using GCC vector extensions in the header for the intrinsic was > a bad idea in this particular case. Yes, the simplest fix is to revert to using the builtin for this particular intrinsic (the builtins haven't been removed, so this is just changing one line in smmintrin.h). Of course this will regress other code, and we are missing out on automatic vectorization opportunities for the same reason, expand should really be fixed (but that's indeed harder). I am a bit surprised we didn't have a testcase in the testsuite to discover that. Although I did manually test most intrinsics when I converted them, I didn't do it with the minimal flags, so there might be one or two others in the same situation as _mm_cmpeq_epi64...
