- * transformation given by ~0, 2, 1, ~5, ~4, ~3. The following sequence of
- * operations is traditional. Essentially this is an antitranspose -- a
- * reflection about the antidiagonal -- followed by a couple of fixup stages.
+ * transformation given by ~0, 2, 1, ~5, ~4, ~3. There's a traditional
+ * swizzle sequence for this, which we used to use, namely:
+ *
+ * // 5, 4, 3, 2, 1, 0
+ * TWIZZLE_XCPL (x, y, 2); // ~2, 4, 3, ~5, 1, 0
+ * SWIZZLE_XCPL2(x, y, 1, 4); // ~2, ~1, 3, ~5, ~4, 0
+ * SWIZZLE_XCPL2(x, y, 0, 3); // ~2, ~1, ~0, ~5, ~4, ~3
+ * SWIZZLE_XCPL2(x, y, 3, 4); // ~2, 0, 1, ~5, ~4, ~3
+ * TWIZZLE_XCPL (x, y, 4); // ~0, 2, 1, ~5, ~4, ~3
+ *
+ * Essentially this is an antitranspose -- a reflection about the
+ * antidiagonal -- followed by a couple of fixup stages. But the non-twizzle
+ * steps require more operations, and it's easy to find a sequence which
+ * always acts on the (current) index bit 5, moving it to where it's wanted,
+ * and inverting it if necessary, so we only need twizzles.