Implicit conversion from float -16.25 to int loses precision