Implicit conversion from float -128.625 to int loses precision