On the computation of the bivariate normal integral