quoted
It'd be great if you could format this patch into a patch set with several
little ones. :-)
Also, the kernel checkpatch is very helpful.
Good coding style and patch organization make it easy for in-depth reviews.
Combination of scalar and vector (32/64/128) was done to get optimal performance numbers. If there is enough interest in this I can work on it and provide an updated patch set.
That'll be very helpful! Looking forward to your patch :)
BTW, have you tested real example performance with your patch?