$ 0 0 Our dense matrices are stored in column-major order, so working by column is faster (i.e. extracting a column vector is a single block copy).