Implicit conversion from float -84.29999999999998 to int loses precision