Abstract: For embedded systems with very limited memory resources, code size becomes one of the most important optimization concerns. Using the oprofile profiling tool, this paper focuses on the Four-Phase Manual Optimization(FPMO) for the software pipelining technique when running the EEMBC benchmark. Experimental result of telecom-autocorrelation shows the FPMO method gets 40.678% performance promotion by increasing 2.04% code size but the pure compiler automatic optimization trades 38.33% performance improvements by 33.35% code size expansion.

Key words: software pipelining, loop unrolling, performance analysis, Four-Phase Manual Optimization(FPMO)

摘要： 代码体积是优化存储资源有限的嵌入式系统的重要因素之一。针对该特点，使用oprofile性能分析工具，以EEMBC基准程序集作为工作负载，提出四阶段人工优化软件流水方法(FPMO)。电信类的自相关程序实验结果表明，FPMO以2.04%的代码增量为代价换来40.678%的性能提升，而单纯的编译器自动优化则以33.35%的体积膨胀换来38.33%的性能提升。

关键词: 软件流水, 循环展开, 性能分析, 四阶段人工优化

CLC Number: 

• TP311