Implicit conversion from float -135.125 to int loses precision