Implicit conversion from float -3.4375 to int loses precision