For floats we would previously only do the division, but not
the truncation for floats. This would result in incorrect values
being returned.