Implicit conversion from float -32.578125 to int loses precision