Implicit conversion from float -16.75 to int loses precision