vxge: prefetch RxD descriptors

This patch prefetches RxD descriptors which helps to lower the latency of a
cache miss in vxge_hw_ring_rxd_next_completed.  This lowers the % of CPU
time used by vxge_hw_ring_rxd_next_completed() where the descriptor is
accessed in profiling netperf on a P4 Xeon from 1.5% to 1.0%.

Signed-off-by: Benjamin LaHaise <ben.lahaise@neterion.com>
Signed-off-by: Sreenivasa Honnur <sreenivasa.honnur@neterion.com>
Signed-off-by: Ramkrishna Vepa <ram.vepa@neterion.com>
Signed-off-by: David S. Miller <davem@davemloft.net>