基于多核平台的程序并行优化研究

发布时间：2018-01-07 20:07

本文关键词：基于多核平台的程序并行优化研究　出处：《长安大学》2012年硕士论文　论文类型：学位论文

【摘要】：随着多核处理器的快速发展，多核并行计算已成为计算机领域的研究热点。对于开发人员来说，由于处理器内核数量的增加与内核功能的变化，原有面向单核或者多处理器的软件架构已不再适于在多核处理器的硬件结构上充分发挥多计算核心的能力，需要相应软件层面的共同发展。多核通用平台开发不同于传统的并行开发在于其平台的通用性、开发周期的快速性和系统级配套辅助工具的使用等方面。因此，如何设计与开发针对多核处理器的应用程序软件，以充分发挥多核处理器带来的巨大计算性能优势，已成为现阶段软件开发中的关键问题。本文主要介绍了多核处理器的发展概况与多核并行编程技术，分析了多核编程与单核编程和多机分布式编程之间的区别，深入研究了在实际应用中迫切需要采用的并行程序设计模式；并利用Intel公司的VTune和Parallel Studio工具对医疗图像数据检查器的串行程序进行分析，找出影响程序性能的“热点”区域，通过对循环并行化、数据保护和线程化等方面的并行优化设计，，结合并行开发工具OpenMP设计出可以并行运行的程序，提出了一种完整且具有良好扩展性的多核并行程序的分析方法和调试步骤；在四核计算平台上获得了理想的加速比，提高了程序的计算性能。验证了在多核计算平台环境下利用并行程序设计理论和并行编程模式相结合开发软件的优越性，达到了对串行程序进行优化的预期目标。
[Abstract]:With the rapid development of multi-core processors, multi-core parallel computing has become a research hotspot in computer field. For developers, due to the increase in the number of processor cores and changes in kernel functions. The original single-core or multi-processor software architecture is no longer suitable for the multi-core processor hardware architecture to give full play to the ability of multi-computing core. Different from the traditional parallel development, the common development of multi-core platform lies in the generality of the platform, the rapidity of the development cycle and the use of auxiliary tools at the system level. How to design and develop the application software for multi-core processors to give full play to the huge computing performance advantages brought by multi-core processors has become a key problem in the software development at this stage. This paper mainly introduces the development of multi-core processor and multi-core parallel programming technology, and analyzes the difference between multi-core programming and single-core programming and multi-computer distributed programming. The concurrent programming pattern which is urgently needed in practical application is deeply studied. The serial program of medical image data checker is analyzed by using VTune and Parallel Studio tools of Intel company, and the "hot spot" area which affects the performance of the program is found out. Through the parallel optimization design of loop parallelization, data protection and threading, combined with the parallel development tool OpenMP, a program that can run in parallel is designed. A complete and extensible analysis method and debugging steps of multi-core parallel programs are presented. The ideal speedup ratio is obtained on the four-core computing platform. The performance of the program is improved, and the superiority of using parallel programming theory and parallel programming mode to develop software is verified under the environment of multi-core computing platform. The expected goal of serial program optimization is achieved.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP332

【参考文献】