摘要
描述了开源高性能编译器Open64及向量化框架,给出了多循环层进行依赖关系分析算法和收益分析方法。结合面向DSP架构的SLP向量化框架,给出了一种应用软件流水的向量化算法。实验结果表明软件流水向量化算法适用于某些计算密集的DSP应用,最高加速比达到14.2。
The open source high performance compiler Open64 and its vectorization framework are described and analyzed. The dependency analysis algorithm and benefit analysis for loop nest are presented. Based on SLP vectorization framework,a software pipelining vectorization algorithm is presented for DSP Architecture. Experimental results show that the software pipelining vectorization algorithm is suitable for compute-intensive DSP applications,with the maximum speedup hitting 14. 2.
出处
《信息工程大学学报》
2015年第6期753-759,共7页
Journal of Information Engineering University
基金
国家863计划资助项目(2009AA01220)
关键词
SLP
循环展开
软件流水
寄存器重用
superword level parallel
loop unrolling
software pipelining
register reuse
