Fix a bug introduced by porting to Vc 1.2
Starting from Vc 1.0 float_v and int_v vectors have different sizes,
so we cannot initialize one from another. And, basically, implicit
initialization in such a way is forbidden in Vc, but the problem is
that we do initialize explicitly.
(cherry picked from commit 56cda65cb7a7b9a56b6694ca27ed3b1fb10cf5ce)