摘要： 在Open64编译框架基础上，提出一种基于Profile信息的循环内数据访问连续性分析算法及其向量化优化方法。采用反馈式编译优化技术，获取程序运行时的连续性Profile信息，通过结构体剥离和数据重组方法实现程序向量化。实验结果表明，该算法针对不规则程序代码，可提供更精确的向量化信息，提高程序的向量化程度。

关键词: 向量化识别, 反馈式编译, 连续性分析, Profile信息, Open64编译框架, 数据重组

Abstract: On the basis of Open64 compiler framework, this paper proposes an algorithm which can implement the continuous reference analysis in nest loops based on profile information and corresponding vectorization optimization method. By using the feed-back compiling optimization techniques, the algorithm can obtain runtime profile information about the continuity of the program and implement loop vectorization by structure peeling and data reorganization. Experimental results show that the algorithm can provide more accurate vectorization information for the irregular code and improve the vectorization extent of code.

Key words: vectorization identification, feed-back compiling, continuity analysis, Profile information, Open64 compiling framework, data reorganization

中图分类号: 

• TP311