摘要
计算划分问题是并行编译中最为重要的问题之一.针对并行循环,在数据分布确定的情况下,提出了基于规范集的计算划分算法,具体讨论了规范集的获取方法及综合通信与负载均衡的最优方案选取算法.实验表明,在并行循环处理方面,这一算法与以前几种算法相比更加简单、有效;采用这一算法的p_HPF编译器对数据并行应用问题可以获得良好的加速比和效率.该编译器已在石油领域得到应用.
Computation partition is one of the most important problems in parallel compilation and optimization. For dealing with parallel loops with determinated data distribution, a computation partition algorithm based on the subset of uniform schemes is proposed. The method of getting the subset of uniform schemes is given, as well as the algorithm of selecting the most optimized scheme under the consideration of communication and load balance. The experimental results prove that this algorithm is simpler and more effective than several previous algorithms in dealing with parallel loops, and the p_HPF compiler adopted by this algorithm can obtain good speedups and efficiencies. The compiler has been applied in the field of petroleum.
出处
《软件学报》
EI
CSCD
北大核心
2003年第3期362-368,共7页
Journal of Software
基金
Supported by the National Natural Science Foundation of China under Grant No.60173004 (国家自然科学基金)
关键词
规范划分集
并行循环
计算划分
并行编译
编译程序
parallel loop
parallel compilation
computation partition
parallel computation
node program
