It's not a new approach; the equivalence was published ~1999 IIRC. It also affects speedup; tensor slicing can now be fully parallelised.