Implicit conversion from float -105.41015625 to int loses precision