Implicit conversion from float -6.796875 to int loses precision