do not differentiate per sample pileups from generic pileups. Do the same for both -- it's O(n) either way.