Application-Level Performance Improvement for Stream Program on CGRA-based systems

Abstract

Coarse-Grained Reconfigurable Architectures (CGRAs), often used as coprocessors for DSP and multimedia kernels, can deliver highly energy-effcient execution for compute-intensive kernels. Simultaneously, stream applications, which consist of many actors and channels connecting them, can provide natural representations for DSP applications, and therefore be a good match for CGRAs. We present our results of mapping DSP applications written in StreamIt language to CGRAs, along with our mapping flow. One important challenge in mapping is how to manage the multitude of kernels in the application for the limited local memory of a CGRA, for which we present a novel integer linear programming-based solution. Our evaluation results demonstrate that our software and hardware optimizations can help generate highly effcient mapping of stream applications to CGRAs, enabling far more energy-effcient executions (7x worse to 50x better) compared to using state-of-theart GP-GPUs. Further, we eliminate communication overhead and reduce computation overhead using combination of sychronous/asynchronous processors and DMA. This optimization also improve performance by 17.1% on average comparing to baseline system.