duffs_device.c

// Tom Duff, 1983 (trivia: for a Lucasfilm animation)

// problem: loop unrolling (reducing the #of loop tests by putting multiple copies of the
//                          given instruction into the body, instead of just one)
// implementation problem: given x-fold unrolling, performing the remainder of iterations,
//                         when count (#of original cycles) is not divisible by x

send(to, from, count)
register short *to, *from;
register count;
{
    register n = (count + 7) / 8;
    switch (count % 8) {
    case 0: do {
              *to = *from++;  // labels 1-7 _inside_ the loop, ie. jumping right into the middle
    case 7:   *to = *from++;  // without `break`, each fall through, till the end
    case 6:   *to = *from++;
    case 5:   *to = *from++;
    case 4:   *to = *from++;
    case 3:   *to = *from++;
    case 2:   *to = *from++;
    case 1:   *to = *from++;
            }
            while (--n > 0);  // still inside the _switch statement_, so the loop runs its course
                              // (now with full cycles) *
    }
}

// * remember, a C `switch` is nothing more than a `goto`: control jumps to
// the statement after the matching label, and then all the rest of the code in
// the block executes (if no `break`)

// Caveat: this is mostly just clever and fun code magic, but performance-wise
// a very-very "it depends" kind of trick, often counterproductive, should be
// considered in rare cases only, if you really know what you're doing
// (https://en.wikipedia.org/wiki/Duff's_device#Performance)