Implicit conversion from float -3.5546875 to int loses precision