It made almost no sense to me when i first saw it, but as with most algorithms, I found that the best way to make sense of it was to step through it in my mind based on a couple of different input scenerios.

In the case of Duff's Device in particular, the real key for me was to keep in mind exactly what it was designed to optimize. If you look at other (more straightforward) loop unwinding optimizations, then this one doesn't seem quite as insane.

(The recent obfu is probably not the best thing to study when trying to make sense of it)