Implicit conversion from float -2.125 to int loses precision