The calculation of pro_super_r rounded to a non-exact float, so put in rounding code for the special cases.