Exploration of Loop Unroll Factors in High Level Synthesis

摘要

The Loop Unrolling optimization can lead to significant performance improvements in High Level Synthesis (HLS), but can adversely affect controller and datapath delays. Unrolling causes additional operations to be scheduled, possibly leading to increased resource sharing, thereby increasing the MUX sizes and delay. We use an estimation based approach to compute the best unroll factor for loops in a behavioral specification by combining a prediction of controller and datapath delays with search space pruning methods. Our exploration strategy gives highly accurate unroll factor results significantly faster than an exhaustive synthesis procedure.