基于多核多线程的混合并行编程技术研究-计算机应用技术专业论文.docxVIP

中原工学院硕士学位论文 中原工学院硕士学位论文 摘要 万方数据 万方数据 基于多核多线程的混合并行编程技术研究 专 业：计算机应用技术 硕 士 生：冉晓龙 指导教师：王文义 教授 摘 要 近十年来，并行处理技术，包括从硬件，系统软件到各类应用软件的发展速度越来越 快，甚至连小型终端的移动手机，处理器等都推出了四核，八核甚至更多，而且还配上了 图形图像处理器 GPU。高性能计算机的不断发展，也极大的促进了各国在许多尖端科技领 域技术的发展。一般情况，在处理具有大量并行任务的作业项目——蛋白质折叠、大气模 拟、地质勘探和能源研究时，才会用到超级并行机。随着大数据和云计算的渐渐兴起和普 及，越来越多的人们开始对并行处理技术发生兴趣。正是这些需求和兴趣，激励人们不断 地去改进、创新和研发新的并行处理软硬件结构和计算方法，进而又不断促进高性能计算 机的发展，可以说供需双方呈现出一种螺旋状的发展态势。 本文首先在单台多核 Linux 平台上，完成了 MPI 环境下不同并行算法的矩阵相乘的性 能对比分析，接着在 GPU 处理器和 CPU 处理器上分别进行了 CUDA 以及多线程 Pthread 矩阵相乘的实验，并且对比分析了两者的实验结果和数据；最后在由两节点组成的最小化 Linux 多核机群上实现了 MPI+Pthread 和 MPI+CUDA 两种混合并行矩阵相乘算法，并分析 了这两种算法所带来的性能优势。在论文的总结部分针对上述实验结果，概括分析了所做 工作的优点、局限性以及今后有待继续改进的努力方向。 关键词：并行处理任务；MPI+Pthread；MPI+CUDA；混合并行编程 I 中原工学院硕士学位论文Abstrac 中原工学院硕士学位论文 Abstract Study on the Hybrid Parallel Programming Technology based on Multi-core and Multi-thread Specialty: Technology of Computer Application Name: Ran Xiaolong Supervisor: Prof. Wang Wenyi Abstract In the past ten years, the development speed of parallel processing technology, including from hardware, system software to various kinds of application software, is becoming more and more quickly. Even the small terminal mobile phone is equipped with a quad-core processor, eight-core or even more. To match with that, some of them are equipped with a graphic processing unit-GPU. The continuous development of high performance has greatly promoted the research progress of many countries in the fields of advanced science and technology. Generally, the super parallel computer can be used only when dealing with a project having a large number of parallel tasks, such as protein folding, atmospheric modeling, geological exploration and energy research. With big data and cloud computing gradually rising and popularizing, more and more people begin to have interest in parallel processing technology. And it is just these needs and interests that guide people to continually improve, innovate, and design new hardware construct and computational method of parallel