Implicit conversion from float -12.25 to int loses precision