Implicit conversion from float -22.500000000000014 to int loses precision