Apply a loop tiling technique to the generic path, which provides performance upside for ISAs with enough registers to take advantage of it. Also helps the compiler optimize this path.
