Implicit conversion from float -145.2 to int loses precision