CUDA编程入门：工程师的高性能并行计算指南

需积分: 10 76 浏览量更新于2024-07-17 3 收藏 6.26MB PDF 举报

《CUDA for Engineers: An Introduction to High-Performance Parallel Computing》是由Duane Storti和Mete Yurtoglu合著的一本面向工程师的入门级书籍，专注于CUDA编程和高性能并行计算。该书于2015年11月由Addison Wesley出版社出版，旨在帮助读者在个人电脑上利用GPU进行高效率的计算，以前这可能需要超级计算机才能完成。本书共331页，适合拥有初级编程经验的科学家、工程师或学生阅读，无需预先了解GPU或并行计算的专业背景。书中采用清晰、简洁的方式讲解CUDA C编程基础，从运行示例程序开始，逐步引导读者编写自己的代码，并通过完整的实例和额外项目加深理解。所有项目都提供详尽的构建说明，适用于主要平台。主要内容包括： 1. **准备计算机**：介绍如何安装和配置硬件及软件环境来运行CUDA程序。 2. **CUDA并行模型与C扩展**：解释CUDA的并行处理模型，以及如何利用C语言进行扩展。 3. **CPU与GPU数据传输**：讨论数据在CPU和GPU之间的传输策略。 4. **性能管理和调试**：涉及时间管理、性能分析、错误处理和调试技术。 5. **2D网格和交互式图形**：学习如何创建2D网格，以及如何与OpenGL结合实现实时用户交互。 6. **差分方程模拟**：演示基本的数值模拟方法。 7. **线程相关的计算**：探讨使用stencil（邻接元素）进行并行处理。 8. **共享内存优化**：利用CUDA的共享内存提高性能。 9. **3D数据交互**：涉及切片、体积渲染和光线投射等操作。 10. **使用CUDA库**：介绍如何利用现成的CUDA库来简化开发过程。 11. **CUDA生态系统探索**：了解更多的CUDA资源和代码库。 12. **实际应用示例**：涵盖了如二维和三维函数可视化、解差分方程、图像处理、线性代数求解、蒙特卡洛计算等多种应用。此外，书中还包含一个附录，对需要时的C编程进行复习，以及关于硬件和软件设置的指南，以及针对CUDA实践中的时间管理、性能评估、错误处理和调试的实用技巧。本书以章节形式展开，共分为9章，从基础知识到实践应用，为读者提供了一个全面的学习CUDA编程路径。无论是希望提升个人电脑计算能力，还是在科研或工程领域应用并行计算技术，这本书都是一个理想的起点。版权信息和目录详细列出了每个章节的内容概要，确保读者能够系统地掌握CUDA技术。

ptg22232966

HOW IS THIS BOOK STRUCTURED?

Appendix B

Software Setup

Appendix D

CUDA Practicalities:

Timing, Proﬁling,

Error Handling,

and Debugging

Appendix A

Hardware Setup

Introduction

Chapter 1

First Steps

Chapter 2

CUDA Essentials

Chapter 3

From Loops to Grids

Chapter 5

Stencils and

Shared Memory

Appendix C

Need-to-Know

C Programming

Chapter 6

Reduction and

Atomic Functions

Chapter 4

2D Grids and

Interactive Graphics

Chapter 9

Exploring the

CUDA Ecosystem

Chapter 7

Interacting with

3D Data

Chapter 8

Using CUDA Libraries

Figure 0.1 Chapter dependencies

ptg22232966

USER’S GUIDE

The codes presented in this book have been created with a high priority on being

concise, readable, explainable, and understandable. As a result, the codes typi-

cally do not include the error-checking and performance-evaluation function ality

that would be present in commercial codes. These topics are covered in Appendix D,

“CUDA Practicalities: Timing, Profiling, Error Handling, and Debugging,” so you

can add those functionalities as desired.

CUDA apps can be developed using either C or C++ programming style. We

primarily use a C programming style to minimize the programming background

required of the reader.

User’s Guide

Our mission in writing CUDA for Engineers is to get a broad technical audience

actively involved in GPU-based parallel computing using CUDA. Thinking meta-

phorically, we are inviting you to take a trip into the land of GPU-based parallel

computing, and CUDA for Engineers is the guidebook for that trip. The travel

guidebook analogy is appropriate in numerous ways, including the following:

• If you only read the guidebook and never actually make the trip, the book has

not really accomplished its intended mission. Please take the trip!

• Most guidebooks deal with things you need to do before starting the trip.

Instead of inoculations and passports, the hardware setup and software

setup required for the trip are covered in Appendices A and B, respectively.

• Travel to foreign lands often involves dealing with a new language and perhaps

a specific dialect. For this trip, the language is C, and the dialect is CUDA. If

you are comfortable with C (or C++), you are in good shape. If C is still foreign

to you, the basics needed to start the trip are covered in Appendix C.

• Guidebooks generally provide a list of must-see places and must-do activities.

Our must-see places are in Chapters 1 and 2, where you will experience the

power of CUDA sample codes and gain initial exposure to the special CUDA

dialect. The must-do activity is in Chapter 3, where you’ll see how to convert

serial C codes into parallel CUDA codes.

• For those desiring the more extensive CUDA tour, just proceed through the

chapters in sequence.

ptg22232966

INTRODUCTION

• Abbreviated and targeted trips are possible, and the departure point is

Chapter 3, where we create our first parallel applications. If you have a sense

of urgency to get to a particular destination, jump right into Chapter 3. If it

makes sense to you, just push forward from there. If not, back up to Chapters 1

and 2 (and/or the appendices) to fill in the gaps as you encounter them. From

there you can proceed in various directions:

• If your goal involves interactive graphics, scientific visualization, gaming,

and the like, proceed to Chapters 4 and 7.

• If your goal is scientific computing, partial differential equations, or image

processing, proceed to Chapters 5 and 6.

• If your primary goal is to utilize existing CUDA-powered libraries, jump

directly to Chapter 8.

• Long tour or short, it is nice to see new things, but actually doing new things

enhances the long-term impact of the experience. So don’t just read. When

codes are presented, create, compile, and execute them. Modify them, and

test out your modifications. Do some of the projects at the end of each chap-

ter, or start up a project of your own.

• On some tours you go solo, and sometimes you travel with a group and a

guide. We have worked hard to produce a book that provides good support for

a solo trip. However, if you end up using the book in a classroom setting, by all

means take advantage of opportunities to interact with instructors and class-

mates. Asking and answering questions is a great way to learn, and a second

pair of eyes is perhaps the best debugging tool.

• On any trip, things may go awry or additional directions or scheduling infor-

mation may be needed. Material for addressing such issues is located in

Appendix D and Chapter 9, which provides pointers to additional resources to

help you start “exploring the CUDA ecosystem.”

Historical Context

This book is the result of several years of adventures into CUDA territory that

started with research projects involving volumetric medical imaging, registra-

tion of 2D and 3D imaging (i.e., fluoroscopy and CT scans), and novel approaches

to computer-aided design and additive manufacturing. As the utility and

ptg22232966

HISTORICAL CONTEXT

importance of CUDA became apparent, it also became clear that sharing the

appreciation of CUDA was the right thing to do. The result was a special topics

course on applications of CUDA offered to students from a wide range of depart-

ments with no prerequisites other than an introductory computing class. The

writing of this book follows the third offering of the course, which has evolved to

include system setup, exploration of CUDA Samples, a very fast introduction to

C programming, CUDA’s model for parallelism—along with the few things

beyond basic C needed to implement it—and a survey of useful CUDA features

including atomic functions, shared memory, and interactive graphics. All of that

happens in the first half of a 10-week quarter, and the remaining time is dedi-

cated to projects and presentations. Each student in the class gives a short pre-

sentation of their project topic in week 5 and then a final presentation in week 10

of what they have accomplished. Class meetings in between are a combination of

consulting hours and presentations (by instructor or guest speakers) on experi-

ences creating and using CUDA-powered applications. In terms of the contents

of the book, the course translates roughly as follows:

• System setup and initial CUDA Sample run: Appendices A and B, Chapter 1.

• Brief introduction to C: Appendix C.

• CUDA, its model for parallelism, and language essentials: Chapters 2 and 3.

• Interactive graphics: Chapter 4.

• Shared memory: First part of Chapter 5.

• Atomic functions: First part of Chapter 6.

• Project: Turn the students loose to explore the remainder of Chapters 5–9

for project inspiration, encourage active writing of code, and provide as much

help as possible getting over hurdles that arise.

While the organization of the book has some relation to the course, the content

is largely the result of a mostly unguided foray into the CUDA world, which turns

out to be a big place where it is not hard to get lost. We have filtered through a

lot of material including books, sample codes, research papers, and seminar

presentations, with the goal of getting down to the gist of what engineers from

outside the field would really need to know to get a handle on what can be done

with a powerful parallel computing tool like CUDA and understand how they can

actually make use of that tool themselves. It is our sincere hope that the result

is a book that can fulfill the dual purposes of enabling individual learning and

剩余330页未读，继续阅读

THESUMMERE

粉丝: 23
资源: 328

CUDA编程入门：工程师的高性能并行计算指南

CUDA for Engineers An Introduction to High-Performance Parallel 无水印原版pdf

CUDA for Engineers An Introduction to High-Performance Parallel Computing

CUDA for Engineers An Introduction to High Performance Parallel Computing

CUDA for Engineers An Introduction to High-Performance Parallel Computing epub

CUDA for Engineers An Introduction to High-Performance Parallel Computing azw3

CUDA11.0-C-Programming-Guide.pdf

cuda_by_example

[Practical Guide]: Building a GAN Model from Scratch: Step-by-Step Optimization for Your First AI ...

【Network Architecture】: Delving into DCGAN and Its Variants: Exploring the Diversity and Potential...

ARCore（Android的增强现实）：ARCore性能优化与调试技巧.docx

最新资源