基于任务的Cholesky分解多核并行化研究  

Research on Cholesky decomposition multi-core parallelism based on task

作　　者：吴华平[1] 郑晓薇[1] 张建强[1] 

机构地区：[1]辽宁师范大学计算机与信息技术学院,辽宁大连116081

出　　处：《计算机工程与设计》2011年第12期4057-4059,4255,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(60603047)

摘　　要：针对Cholesky分解算法采用OpenMP并行程序设计时的并行性开销增大和线程负载不平衡的问题,利用并行性能分析工具对串行程序进行热点分析,提出了一种基于任务的Cholesky分解多核并行算法。该算法将大循环问题划分成各个相互独立的小任务,并运用任务窃取技术和动态负载均衡算法使多个任务能够并行完成。采用ParallelAmplifier对并行程序进行调试和优化,实验结果表明,其性能得到较大幅度的提升。In order to solve the problem that the cost increases and the thread load is imbalance when we adopt OpenMP parallel programming for Cholesky decomposition algorithm.A parallel algorithm is proposed,which is a task-based multi-core for Cholesky decomposition and a serial programs are analyzed using the parallel studio tools.This algorithm divides the large circulation problems into each independent small tasks,using the task stealing technique and dynamic load balancing algorithm to make multiple tasks to be parallel.The Parallel Amplifier is used to debug and optimize the parallel programs.And the experiment shows that the performance is improved substantially.

关 键 词：CHOLESKY分解 并行化 热点分析 任务窃取技术 动态负载均衡 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]