GitLab

division instead of RCPPS-followed-by-Newton-Raphson. The rationale for that is
that elsewhere in Eigen we dont allow ourselves this approximation (which throws
2 bits of mantissa), so there's no reason we should allow it here.