Implicit conversion from float -12.799999999999997 to int loses precision