Implicit conversion from float -32.34375 to int loses precision