Implicit conversion from float -20.578512396694222 to int loses precision