For example, one may want to split sparse matrices into matrices with just 1M rows, and do the the (top-n) multiplication of all those matrix pairs. Reasons to do this are to reduce the memory ...