https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125880
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
When AVX2 is available vpbroadcast{b,w} from memory might be another option,
removing the necessity of the xorps (so better for code size). This
works only when all other remaininig elements of the vector are overwritten
(like it is with vec_init). Watch out for sub-SSE vector modes, of course.
