Right now, HVPs are computed individually and assembled, but this was only useful for the sparse case
Right now, HVPs are computed individually and assembled, but this was only useful for the sparse case