This change allows us to save memory on loading, by skipping one copy
step. Also loading is a bit faster. Next-up is 1D row vector loading.