Implicit conversion from float -123.28125 to int loses precision