https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116493
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note WIDEN_SUM_EXPR is listed as lane-reducing when applied to
vectors (tree.def isn't clear and of course generic.texi doesn't document it).
So I don't think the proposed simplification works. It would work
for { 0, 0, 0, 0 } w+ vect__4.6_1, simplifying to vect__4.6_1 though
(the op isn't commutative either).
