Implicit conversion from float -8.5 to int loses precision