Implicit conversion from float -4.976635514018682 to int loses precision