Implicit conversion from float -3.21608040201005 to int loses precision