Returns
a vector containing the largest representable floating-point integral
values less than or equal to the values of the corresponding elements
of the given vector.
Note: vec_roundm provides
the same functionality as vec_floor, except that vec_roundm would
not trigger the IEEE-inexact exception.