udiv_qrnnd(q, r, nh, nl, d)

   Core2 53.4 cycles
   K8 74.3 cycles
   K10 80.6 cycles

udiv_qrnnd_preinv(q, r, nh, nl, d, di)

   Core2 21.7 cycles
   K8 14.9 cycles
   K10 17.0 cycles
