Shift-out skipped mask bits for both crpred and intpred

Shift-out skipped mask bits for both crpred and intpred

If src/dest step are not zero, we need to shift-out the skipped mask
bits. We already did this for intpred, and for crpred it's exactly the
same.
Move the shifting logic to a new last state, commonly used for both
intpred and crpred.