Doing it fast and accurately is the hard part. This paper touches on some of the numerical accuracy improvements they made to the algorithm to help out. It involves some rather large numbers, so numerical accuracy of the floating point calculations matters.
