Implicit conversion from float -9.25 to int loses precision