Implicit conversion from float -2.581395348837219 to int loses precision