Implicit conversion from float -57.5 to int loses precision