This is the same method as in pull request #455, but using integer
arithmetic instead of converting to float.

Fixes #455.