Implicit conversion from float -21.038961038961034 to int loses precision