Implicit conversion from float -112.5 to int loses precision