Okay, but what if you're dealing with a rather high-dimensional tensor? In some kinds of coding it can happen, and you usually don't want to sacrifice performance when it does.

You can also do increasingly elaborate pointer arithmetic, but that seems worse, not better to me.