https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123266
--- Comment #2 from Victor Do Nascimento <victorldn at gcc dot gnu.org> --- Yes, I'm afraid I see your point. Having said that, however, I also realize that as things stand, our value range propagation pass doesn't track ranges for vectors. Of course, this is somewhat understandable given that most cases involving vectors won't have uniform value vectors such as here. As such, what we see here (and as you kindly point out in the bug title), my change doesn't so much introduce the bug as it exposes an existing bug resulting from the limitations to the value range propagation to vec_duplicate_expr vectors. A fix will involve fairly big changes extending VRP to vector types and certainly beyond what we can reasonably expect to accomplish in time for the GCC16 release (plus it'd be way too invasive).
