Implicit conversion from float -48.5625 to int loses precision