Implicit conversion from float -4.53125 to int loses precision