Performance Visualization and Exploration for Reconfigurable Computing Applications

Reconfigurable computing (RC) applications have the potential for significant performance while consuming little power. Although runtime performance analysis of RC applications has previously been shown to be important in achieving this potential, the optimization process is still arduous. We target two primary contributing factors: the incongruence between traditional CPU-based performance visualizations and RC performance data, and the need to manually predict the benefits of each potential optimization. To address these issues, we propose a methodology for interactive performance visualization targeted toward RC applications. Our methodology presents performance data hierarchically within the system and application contexts while simultaneously permitting a user to explore “what-if” scenarios to predict the effects of optimizations. We present a prototype of this visualization in our ReCAP (Reconfigurable Computing Application Performance) tool and demonstrate its utility via an application case study.