Automatic Customization of Embedded Applications for Enhanced Performance and Reduced Power Using Optimizing Compiler Techniques

摘要

This paper introduces a compiler framework that optimizes embedded applications written in C, and produces high-level hardware descriptions of the applications for customization on Field-Programmable Gate Arrays (FPGAs). Our compiler performs machine-specific and machine-independent optimizations in order to increase the performance of an embedded application and reduce area/power requirements without explicit programmer intervention. Our experimental results show that our compiler framework can increase performance by 38% with loop and expression parallelism for eight embedded benchmarks. Also, area usage and power consumption are reduced by 69% and 55%, respectively through the efficient utilization of on-chip FPGA resources for Xilinx Virtex-II FPGA chip.