Implicit conversion from float -40.6640625 to int loses precision