Implicit conversion from float -2.513513513513516 to int loses precision