Implicit conversion from float -9.21875 to int loses precision