Implicit conversion from float -31.621621621621614 to int loses precision