Niels, > And if I understood the paper correctly, the growth is slightly worse, > since the right-to-left Jacobi algorithm uses non-negative quotients, > while the corresponding gcd algorithm works with signed quotients. > Right? yes. Paul