Cost-Driven Hybrid Configuration Prefetching for Partial Reconfigurable Coprocessor

Reconfigurable computing systems have developed the capability of changing the configuration of the reconfigurable coprocessor multiple times during the course of a program. However, in most systems the reconfigurable coprocessor wastes computation cycles while waiting for the reconfiguration to complete. Therefore, the high demand for frequent run-time reconfiguration directly translates into higher reconfiguration overhead. Some studies have introduced the concept of prefetching to reduce the reconfiguration overhead. However, these prefetching algorithms are probability-driven. We believe that including configuration size information in the prediction algorithm directly links the training of the predictor with the performance gain. Therefore we proposed a performance-oriented cost-driven algorithm for coarse-grained configuration prefetching. Our cycle accurate simulation results show that the proposed cost-driven algorithm outperforms the probability-driven predictor by 10.8% to 29.6% in reducing reconfiguration overhead.