Implicit conversion from float -12.34375 to int loses precision