Implicit conversion from float -2.756756756756772 to int loses precision