Implicit conversion from float -33.515625 to int loses precision