Implicit conversion from float -21.643835616438352 to int loses precision