OpenMP API指南：并行编程核心技术 - CSDN文库

4星 · 超过85%的资源需积分: 10 160 浏览量更新于2024-08-02 1 收藏 1.99MB PDF 举报

OpenMP (Open Multi-Processing) 是一种并行编程接口，用于在共享内存系统上实现高效的多线程编程。这份文档提供了OpenMP应用接口的详细指南，由四位来自IT领域的专家共同编撰和贡献：Rohit Chandra, Leonardo Dagum, Dave Kohr, 和 Dror Maydan。Rohit Chandra作为NARUS公司的首席科学家，曾在Silicon Graphics（SGI）的编译器组工作，参与了OpenMP的设计与实现，他在此领域有着深厚的技术背景。Leonardo Dagum则在SGI的Linux Server Platform Group任职，负责Scalable Linux服务器系统的I/O基础设施，他在OpenMP Fortran API的定义中发挥了关键作用，研究兴趣包括并行算法和并行系统性能建模。 Dave Kohr作为NARUS公司的技术人员，同样在SGI的编译器组有过工作经历，对OpenMP的定义和实现有所贡献。Dror Maydan在Tensilica公司担任软件总监，曾是SGI编译器组的工程部门经理，他的专业知识对于OpenMP的设计和实践起到了推动作用。Jeff McDonald则在SolidFX这家私人软件开发公司任职，作为工程部门经理，他可能在OpenMP的实际应用和优化方面有着丰富的实践经验。本文件涵盖了OpenMP的主要概念，如并行编程的基本原理、如何在OpenMP环境中组织和管理线程、同步机制（如barrier和critical sections）、数据共享和同步以及错误处理等核心话题。此外，还可能探讨了OpenMP在不同编程语言（如C和Fortran）中的应用示例和最佳实践，以及如何利用OpenMP进行任务并行化、工作分配和负载平衡，以提高程序的执行效率。学习者可以通过这份手册深入了解如何利用OpenMP进行高性能计算，无论是对于科研人员还是软件工程师，它都是一个宝贵的资源，帮助他们简化并行编程过程，提升代码的可维护性和性能。

Preface xv

Chapter 5, “Synchronization,” describes the synchronization mecha-

nisms in OpenMP. It describes the situations in a shared memory parallel

program when explicit synchronization is necessary. It presents the vari-

ous OpenMP synchronization constructs and also describes how program-

mers may build their own custom synchronization in a shared memory

parallel program.

Chapter 6, “Performance,” discusses the performance issues that arise

in shared memory parallel programs. The only reason to write an OpenMP

program is scalable performance, and this chapter is a must-read to realize

these performance goals.

Appendix A, A Quick Reference to OpenMP, which details various

OpenMP directives, runtime library routines, lock routines, and so on, can

be found immediately following Chapter 6.

A presentation note regarding the material in the text. The code frag-

ments in the examples are presented using a different font, shown below:

This is a code sample in an example.

This is an OpenMP construct in an example.

Within the examples all OpenMP constructs are highlighted in bold-

face monofont. Code segments such as variable names, or OpenMP con-

structs used within the text are simply highlighted using the regular text

font, but in italics as in this sample.

Acknowledgments

This book is based entirely on the OpenMP effort, which would not have

been possible without the members of the OpenMP Architectural Review

Board who participated in the design of the OpenMP speciﬁcation, includ-

ing the following organizations: Compaq Computer Corporation, Hewlett-

Packard Company, Intel Corporation, International Business Machines

(IBM), Kuck & Associates (KAI), Silicon Graphics, Inc. (SGI), Sun Micro-

systems, Inc., and the U.S. Department of Energy ASCI Program.

This book beneﬁted tremendously from the detailed perusal of several

reviewers, including George Adams of Purdue University, Tim Mattson of

Intel Corporation, David Kuck of Kuck & Associates, Ed Rothberg of Ilog,

Inc. (CPLEX Division), and Mary E. Zosel of the U.S. Department of Energy

ASCI Program.

2 Chapter 1—Introduction

target architecture is described at a high level, followed by a brief explana-

tion of why OpenMP was developed and how it came to be. Finally, a road

map of the remainder of the book is presented to help navigate through

the rest of the text. This will help readers with different levels of experi-

ence in parallel programming come up to speed on OpenMP as quickly as

possible.

This book assumes that the reader is familiar with general algorithm

development and programming methods. No speciﬁc experience in paral-

lel programming is assumed, so experienced parallel programmers will

want to use the road map provided in this chapter to skip over some of the

more introductory-level material. Knowledge of the Fortran language is

somewhat helpful, as examples are primarily presented in this language.

However, most programmers will probably easily understand them. In

addition, there are several examples presented in C and C++ as well.

1.1

Performance with OpenMP

Applications that rely on the power of more than a single processor are

numerous. Often they provide results that are time-critical in one way or

another. Consider the example of weather forecasting. What use would a

highly accurate weather model for tomorrow’s forecast be if the required

computation takes until the following day to complete? The computational

complexity of a weather problem is directly related to the accuracy and

detail of the requested forecast simulation. Figure 1.1 illustrates the perfor-

mance of the MM5 (mesoscale model) weather code when implemented

using OpenMP [GDS 95] on an SGI Origin 2000 multiprocessor.

1

The graph

shows how much faster the problem can be solved when using multiple

processors: the forecast can be generated 70 times faster by using 128 pro-

cessors compared to using only a single processor. The factor by which the

time to solution can be improved compared to using only a single proces-

sor is called speedup.

These performance levels cannot be supported by a single-processor

system. Even the fastest single processor available today, the Fujitsu

VPP5000, which can perform at a peak of 1512 Mﬂop/sec, would deliver

MM5 performance equivalent to only about 10 of the Origin 2000 proces-

sors demonstrated in the results shown in Figure 1.1.

2

Because of these

1 MM5 was developed by and is copyrighted by the Pennsylvania State University (Penn State)

and the University Corporation for Atmospheric Research (UCAR). MM5 results on SGI

Origin 2000 courtesy of Wesley Jones, SGI.

2 http://box.mmm.ucar.edu/mm5/mpp/helpdesk/20000106.html.

剩余248页未读，继续阅读

yongzhou304011

粉丝: 0
资源: 1

最新资源