Implicit conversion from float -55.625 to int loses precision