The 64b addition operation is broken into two 32b additions (with carry-in). This allows a 64b addition to be completed every cycle, but requires an additional cycle of latency on each addition.