Implicit conversion from float -5.3125 to int loses precision