摘要： 计算和数据分解是分布主存系统中并行编译的关键，在并行优化编译器的并行识别过程中，许多串行代码无法找到全局一致的分解结果。针对这种情况，该文提出一种融合程序控制流的动态分解算法，增加控制流对分解的影响，使生成的分解结果更适合于后端自动生成的并行代码。实验分析结果表明了该方法的有效性。

关键词: 并行编译, 动态分解, 数据重分布, 控制流

Abstract: Computation and data decompositions are key factors of parallel compiler on distributed memory parallel computers. In the process of parallel recognition, many serial codes have no way to find a uniform decomposition result. This paper presents a dynamic decomposition algorithm based on merging control flow analysis. This method, based on the former researches, enhances the influence of control flow to the decomposition and makes the result more suitable for the execution of parallel codes generated by end. Experimental analysis shows the method is effective.

Key words: parallel compiler, dynamic decomposition, data reorganization, control flow

中图分类号: 

• TP393