We need to shift only one time in each cycle in the swapping loop
for unaligned tx case. Currently two byte shift operations are
performed in each loop cycle causing zero gaps in the transmited
data, so not all data scheduled for transmition is actually
transmited.

The proper swapping in unaligned rx case is missing, so add it
as we need to put the received data into the rx buffer in the
correct byte order.

Signed-off-by: Anatolij Gustschin <agust@denx.de>
Tested-by: Stefano Babic <sbabic@denx.de>