Implicit conversion from float -5.5078125 to int loses precision