图书介绍

大规模并行处理器程序设计2025|PDF|Epub|mobi|kindle电子书版本百度云盘下载

（美）柯克（Kirk.D.）著著
出版社：北京市：清华大学出版社
ISBN：9787302229735
出版时间：2010
标注页数：258页
文件大小：59MB
文件页数：278页
主题词：并行程序－程序设计－高等学校－教材－英文

PDF下载

点此进入-本书在线PDF格式电子书下载【推荐-云解压-方便快捷】直接下载PDF格式图书。移动端-PC端通用
种子下载[BT下载速度快]温馨提示：（请使用BT下载软件FDM进行下载）软件下载地址页直链下载[便捷但速度慢] [在线试读本书] [在线获取解压码]

点击复制MD5值：bd07a3dba77008f1f60bcb5ea02c8858

下载说明

大规模并行处理器程序设计PDF格式电子书版下载

下载的文件为RAR压缩包。需要使用解压软件进行解压得到PDF格式图书。

点击复制85GB完整离线版磁力链接到迅雷FDM等BT下载工具进行下载详情点击-查看共享计划

建议使用BT下载工具Free Download Manager进行下载,简称FDM(免费,没有广告,支持多平台）。本站资源全部打包为BT种子。所以需要使用专业的BT下载软件进行下载。如BitComet qBittorrent uTorrent等BT下载工具。迅雷目前由于本站不是热门资源。不推荐使用！后期资源热门了。安装了迅雷也可以迅雷进行下载！

（文件页数要大于标注页数，上中下等多册电子书除外）

注意：本站所有压缩包均有解压码： 点击下载压缩包解压工具

图书目录

CHAPTER 1 INTRODUCTION1

1.1 GPUs as Parallel Computers2

1.2 Architecture of a Modern GPU8

1.3 Why More Speed or Parallelism?10

1.4 Parallel Programming Languages and Models13

1.5 Overarching Goals15

1.6 Organization of the Book16

CHAPTER 2 HISTORY OF GPU COMPUTING21

2.1 Evolution of Graphics Pipelines21

2.1.1 The Era of Fixed-Function Graphics Pipelines22

2.1.2 Evolution of Programmable Real-Time Graphics26

2.1.3 Unified Graphics and Computing Processors29

2.1.4 GPGPU:An Intermediate Step31

2.2 GPU Computing32

2.2.1 Scalable GPUs33

2.2.2 Recent Developments34

2.3 Future Trends34

CHAPTER 3 INTRODUCTION TO CUDA39

3.1 Data Parallelism39

3.2 CUDA Program Structure41

3.3 A Matrix-Matrix Multiplication Example42

3.4 Device Memories and Data Transfer46

3.5 Kernel Functions and Threading51

3.6 Summary56

3.6.1 Function declarations56

3.6.2 Kernel launch56

3.6.3 Predefined variables56

3.6.4 Runtime API57

CHAPTER 4 CUDA THREADS59

4.1 CUDA Thread Organization59

4.2 Using blockIdx and threadIdx64

4.3 Synchronization and Transparent Scalability68

4.4 Thread Assignment70

4.5 Thread Scheduling and Latency Tolerance71

4.6 Summary74

4.7 Exercises74

CHAPTER 5 CUDATM MEMORIES77

5.1 Importance of Memory Access Efficiency78

5.2 CUDA Device Memory Types79

5.3 A Strategy for Reducing Global Memory Traffic83

5.4 Memory as a Limiting Factor to Parallelism90

5.5 Summary92

5.6 Exercises93

CHAPTER 6 PERFORMANCE CONSIDERATIONS95

6.1 More on Thread Execution96

6.2 Global Memory Bandwidth103

6.3 Dynamic Partitioning of SM Resources111

6.4 Data Prefetching113

6.5 Instruction Mix115

6.6 Thread Granularity116

6.7 Measured Performance and Summary118

6.8 Exercises120

CHAPTER 7 FLOATING POINT CONSIDERATIONS125

7.1 Floating-Point Format126

7.1.1 Normalized Representation of M126

7.1.2 Excess Encoding of E127

7.2 Representable Numbers129

7.3 Special Bit Patterns and Precision134

7.4 Arithmetic Accuracy and Rounding135

7.5 Algorithm Considerations136

7.6 Summary138

7.7 Exercises138

CHAPTER 8 APPLICATION CASE STUDY:ADVANCED MRI RECONSTRUCTION141

8.1 Application Background142

8.2 Iterative Reconstruction144

8.3 Computing FHd148

Step 1．Determine the Kernel Parallelism Structure149

Step 2．Getting Around the Memory Bandwidth Limitation156

Step 3．Using Hardware Trigonometry Functions163

Step 4．Experimental Performance Tuning166

8.4 Final Evaluation167

8.5 Exercises170

CHAPTER 9 APPLICATION CASE STUDY:MOLECULAR VISUALIZATION AND ANALYSIS173

9.1 Application Background174

9.2 A Simple Kernel Implementation176

9.3 Instruction Execution Efficiency180

9.4 Memory Coalescing182

9.5 Additional Performance Comparisons185

9.6 Using Multiple GPUs187

9.7 Exercises188

CHAPTER 10 PARALLEL PROGRAMMING AND COMPUTATIONAL THINKING191

10.1 Goals of Parallel Programming192

10.2 Problem Decomposition193

10.3 Algorithm Selection196

10.4 Computational Thinking202

10.5 Exercises204

CHAPTER 11 A BRIEF INTRODUCTION TO OPENCLTM205

11.1 Background205

11.2 Data Parallelism Model207

11.3 Device Architecture209

11.4 Kernel Functions211

11.5 Device Management and Kernel Launch212

11.6 Electrostatic Potential Map in OpenCL214

11.7 Summary219

11.8 Exercises220

CHAPTER 12 CONCLUSION AND FUTURE OUTLOOK221

12.1 Goals Revisited221

12.2 Memory Architecture Evolution223

12.2.1 Large Virtual and Physical Address Spaces223

12.2.2 Unified Device Memory Space224

12.2.3 Configurable Caching and Scratch Pad225

12.2.4 Enhanced Atomic Operations226

12.2.5 Enhanced Global Memory Access226

12.3 Kernel Execution Control Evolution227

12.3.1 Function Calls within Kernel Functions227

12.3.2 Exception Handling in Kernel Functions227

12.3.3 Simultaneous Execution of Multiple Kernels228

12.3.4 Interruptible Kernels228

12.4 Core Performance229

12.4.1 Double-Precision Speed229

12.4.2 Better Control Flow Efficiency229

12.5 Programming Environment230

12.6 A Bright Outlook230

APPENDIX A MATRIX MULTIPLICATION HOST-ONLY VERSION SOURCE CODE233

A.1 matrixmul.cu233

A.2 matrixmul_gold.cpp237

A.3 matrixmul.h238

A.4 assist.h239

A.5 Expected Output243

APPENDIX B GPU COMPUTE CAPABILITIES245

B.1 GPU Compute Capability Tables245

B.2 Memory Coalescing Variations246

Index251