This is not 8 times faster than byte access, but still around 60% faster.
Signed-off-by: NNicolas Pitre <nico@marvell.com>
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>