Save all visited sums and corresponding indexes in a priority queue. Then, once you pop the smallest sum so far, you can quickly identify the next m candidates for smallest sum by incrementing each row index by 1.