Implicit conversion from float -1.678622668579635 to int loses precision