// [ 8 9 10 11] [ 8 13 2 7] (c, q10)
// [12 13 14 15] [12 1 6 11] (d, q11)
//
- // [ 0 1 2 3] (a, q8)
- // [ 4 5 6 7] (b, q9)
- // [ 8 9 10 11] (c, q10)
- // [12 13 14 15] (d, q11)
- //
// We need a copy for later. Rather than waste time copying them by
// hand, we'll use the three-address nature of the instruction set.
// But this means that the main loop is offset by a bit.