Duane Storti is a professor of mechanical engineering at the University of Washington in Seattle. He has thirty-five years of experience in teaching and research in the areas of engineering mathematics, dynamics and vibrations, computer-aided design, 3D printing, and applied GPU computing.

Mete Yurtoglu is currently pursuing an M.S. in applied mathematics and a Ph.D. in mechanical engineering at the University of Washington in Seattle. His research interests include GPU-based methods for computer vision and machine learning.

Acknowledgments xvii

About the Authors xix

Introduction 1

What Is CUDA? 1

What Does “Need-to-Know” Mean for Learning CUDA? 2

What Is Meant by “for Engineers”? 3

What Do You Need to Get Started with CUDA? 4

How Is This Book Structured? 4

Conventions Used in This Book 8

Code Used in This Book 8

User’s Guide 9

Historical Context 10

References 12

Chapter 1: First Steps 13

Running CUDA Samples 13

Running Our Own Serial Apps 19

Summary 22

Suggested Projects 23

Chapter 2: CUDA Essentials 25

CUDA’s Model for Parallelism 25

Need-to-Know CUDA API and C Language Extensions 28

Summary 31

Suggested Projects 31

References 31

Chapter 3: From Loops to Grids 33

Parallelizing dist_v1 33

Parallelizing dist_v2 38

Standard Workflow 42

Simplified Workflow 43

Summary 47

Suggested Projects 48

References 48

Chapter 4: 2D Grids and Interactive Graphics 49

Launching 2D Computational Grids 50

Live Display via Graphics Interop 56

Application: Stability 66

Summary 76

Suggested Projects 76

References 77

Chapter 5: Stencils and Shared Memory 79

Thread Interdependence 80

Computing Derivatives on a 1D Grid 81

Summary 117

Suggested Projects 118

References 119

Chapter 6: Reduction and Atomic Functions 121

Threads Interacting Globally 121

Implementing parallel_dot 123

Computing Integral Properties: centroid_2d 130

Summary 138

Suggested Projects 138

References 138

Chapter 7: Interacting with 3D Data 141

Launching 3D Computational Grids: dist_3d 144

Viewing and Interacting with 3D Data: vis_3d 146

Summary 171

Suggested Projects 171

References 171

Chapter 8: Using CUDA Libraries 173

Custom versus Off-the-Shelf 173

Thrust 175

cuRAND 190

NPP 193

Linear Algebra Using cuSOLVER and cuBLAS . 201

cuDNN 207

ArrayFire 207

Summary 207

Suggested 208

References 209

Chapter 9: Exploring the CUDA Ecosystem 211

The Go-To List of Primary Sources 211

Further Sources 217

Summary 218

Suggested Projects 219

Appendix A: Hardware Setup 221

Checking for an NVIDIA GPU: Windows 221

Checking for an NVIDIA GPU: OS X 222

Checking for an NVIDIA GPU: Linux 223

Determining Compute Capability 223

Upgrading Compute Capability 225

Appendix B: Software Setup 229

Windows Setup 229

OS X Setup 238

Linux Setup 240

Appendix C: Need-to-Know C Programming 245

Characterization of C 245

C Language Basics 246

Data Types, Declarations, and Assignments 248

Defining Functions 250

Building Apps: Create, Compile, Run, Debug 251

Arrays, Memory Allocation, and Pointers 262

Control Statements: for, if 263

Sample C Programs 267

References 277

Appendix D: CUDA Practicalities: Timing, Profiling, Error Handling, and Debugging 279

Execution Timing and Profiling 279

Error Handling 292

Debugging in Windows 298

Debugging in Linux 305

CUDA-MEMCHECK 308

Using Visual Studio Property Pages 309

References 312

Index 313