CASIA OpenIR  > 毕业生  > 硕士学位论文
基于MaPU的异构多核编程技术研究
其他题名The study of Heterogeneous Multicore Programming Techniques based on MaPU
林建海
学位类型工程硕士
导师尹志刚
2014-05-23
学位授予单位中国科学院大学
学位授予地点中国科学院自动化研究所
学位专业计算机应用技术
关键词Mapu Sift 异构多核处理器 并行技术 暂态稳定分析 Mapu Sift Heterogeneous Multicore Processors Parallel Technology Transient Stability Analysis
摘要异构多核处理器不断普及给程序员进行软件编程设计带来了前所未有的挑战。在市面上应用程序的更新速度快,要求软件设计必须要有较短的周期性。然而若直接将同构多核并行程序移植到异构多核处理器上运行,则不能够充分利用异构多核处理器的结构优势,因而需要重新编写适应于异构多核处理器的并行程序。异构多核处理器在提升软件应用性能的同时也加大了软件编程的难度,因此人们有必要研究如何提高异构多核应用程序的设计效率。 本文主要在MaPU(Mathematic Processing Unit)高性能代数处理器的基础上,研究基于异构多核处理器的编程技术。本文通过对多种计算密集型算法进行深入研究,并针对异构多核处理器的架构特点提出了一种基于异构多核处理器的并行设计模型,本文称之为CCD(Control Calculation Divide)模型。该模型的主要思想是将算法中的控制部分和计算部分分离,将控制部分放在主控核中运行,计算部分经过多核分解后放在计算核中运行。同时,本文在此基础上也提出了一种基于任务队列的半动态任务调度方式,将该调度方式分成可完全并行的任务调度和任务间有相关性的任务调度两种情况,并分别阐述了这两种情况的实施方案。 本文最后将本文提出的基于异构多核处理器的CCD模型和基于任务队列的半动态任务调度方式应用在两个实际工程应用中。首先,利用CCD模型,在基于MaPU的超级计算平台上,对电网安全分析中的潮流计算和暂态分析算法中的最主要的牛顿迭代过程进行并行化分解设计,并对其性能进行评估。最后,将CCD模型和基于任务队列的半动态任务调度方法应用于SIFT特征点检测算法的并行化设计与实现中,最终实验表明本文的设计模型和任务调度方式在加速比和负载均衡等方面有很好的性能指标。 本文的研究在后续对基于MaPU处理器的软件设计上具有一定的指导性意义,甚至在对基于其他异构多核处理器的软件设计也具有一定的指导性意义。
其他摘要The growing popularity of heterogeneous multicore processors brings unprecedented challenges to the programmer programming software. In the market, the speed of updating application is faster, which requires software design must have shorter periodicity. However, if the homogeneous multicore parallel programs are directly ported to heterogeneous multicore processors, it won’t make full use of the advantages of heterogeneous multicore processors on architecture, and therefore we need to be rewritten to adapt to heterogeneous multicore processors. Heterogeneous multicore processors improved the performance of software also increased the difficulty of software programming, so people need to study how to improve the design efficiency of heterogeneous multicore processors. The dissertation mainly studies the programming of heterogeneous multicore processors technology based on MaPU. Through in-depth studies on several computationally intensive algorithms, and based on the architectural features of the heterogeneous multicore processors, the dissertation proposes a parallel design model based on heterogeneous multicore processors, in the dissertation we name it CCD model. The main idea of this model is to divide the algorithms into the control portion and the calculating portion. Then, the control portion runs on the master core, and the calculating portion runs on the slave core after decomposing. The dissertation focuses on the achieving of multicore decomposition and programming. Meanwhile, the dissertation also proposes a semi-dynamic task scheduling based on task queue. And this task scheduling is divided into two cases in the dissertation. One case is the task can fully parallel and the other case is there is correlation between tasks. The embodiments are explained in both cases. In the dissertation, the CCD model based on heterogeneous multicore processors, and the semi-dynamic task scheduling based on task queue are applied to two practical engineering applications. Firstly, using CCD model to decompose the Newton iterative algorithm of flow calculation and transient analysis in power system security analysis, based on MaPU supercomputing platform. And evaluate its performance. Finally, we use the CCD model and the semi-dynamic task scheduling to design and implement the parallel algorithm of SIFT. The experimental results show that the CCD model and the semi-dynamic task scheduling of this dissertation has good performance in terms of speedup and l...
馆藏号XWLW2072
其他标识符2011E8014661095
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/7704
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
林建海. 基于MaPU的异构多核编程技术研究[D]. 中国科学院自动化研究所. 中国科学院大学,2014.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_2011E801466109(2027KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[林建海]的文章
百度学术
百度学术中相似的文章
[林建海]的文章
必应学术
必应学术中相似的文章
[林建海]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。