Implicit conversion from float -21.621621621621614 to int loses precision