Implicit conversion from float -21.548731642189594 to int loses precision