摘要
为了在多核处理器上充分利用多核资源以提升程序性能,研究了多核处理器的体系结构和多核环境下可能影响并行程序性能的因素,实现了基于任务的并行编程模型。该模型提供了单任务数据并行和多任务并行两种并行处理方式,其中单任务数据并行使用cache块技术划分数据集,多任务并行使用任务密取的任务调度策略。用该模型实现了计算斐波那契数列的递归算法,实验结果表明,使用该模型编写多核并行程序可以达到较高的相对于串行计算的加速比。
To fully utilize the multi-core resources on multi-core processors to improve program performance,a parallel programming model based on tasks is implemented with the study of the system architecture of multi-core processor and the factors that may influence the performance of parallel programs.One-task data parallelism and multi-task parallelism are provided by this model.Cache blocking technique is used to divide data set in one-task data parallelism and working stealing is used to schedule tasks in multi-task parallelism.The recursive algorithm of Fibonacci is realized by this model and the experimental result indicates that multi-core parallel programs written by this model can reach high speedup relative to serial programs.
出处
《计算机工程与设计》
CSCD
北大核心
2010年第13期2999-3002,3056,共5页
Computer Engineering and Design
基金
上海市重点学科建设基金项目(J50103)