Programming Massively Parallel Processors

A Hands-on Approach

3rd Edition - December 7, 2016
Authors: David B. Kirk, Wen-mei W. Hwu
Language: English
Paperback ISBN:
9 7 8 - 0 - 1 2 - 8 1 1 9 8 6 - 0
eBook ISBN:
9 7 8 - 0 - 1 2 - 8 1 1 9 8 7 - 7

Programming Massively Parallel Processors: A Hands-on Approach, Third Edition shows both student and professional alike the basic concepts of parallel programming and GPU architect… Read more

Purchase options

BACK-TO-SCHOOL

Fuel your confidence!

Up to 25% off learning resources

Shop now

Programming Massively Parallel Processors: A Hands-on Approach, Third Edition shows both student and professional alike the basic concepts of parallel programming and GPU architecture, exploring, in detail, various techniques for constructing parallel programs.

Case studies demonstrate the development process, detailing computational thinking and ending with effective and efficient parallel programs. Topics of performance, floating-point format, parallel patterns, and dynamic parallelism are covered in-depth.

For this new edition, the authors have updated their coverage of CUDA, including coverage of newer libraries, such as CuDNN, moved content that has become less important to appendices, added two new chapters on parallel patterns, and updated case studies to reflect current industry practices.

Dedication
Preface
- Target Audience
- How to Use the Book
- Illinois–NVIDIA GPU Teaching Kit
- Online Supplements
Acknowledgements
Chapter 1. Introduction
- Abstract
- 1.1 Heterogeneous Parallel Computing
- 1.2 Architecture of a Modern GPU
- 1.3 Why More Speed or Parallelism?
- 1.4 Speeding Up Real Applications
- 1.5 Challenges in Parallel Programming
- 1.6 Parallel Programming Languages and Models
- 1.7 Overarching Goals
- 1.8 Organization of the Book
- References
Chapter 2. Data parallel computing
- Abstract
- 2.1 Data Parallelism
- 2.2 CUDA C Program Structure
- 2.3 A Vector Addition Kernel
- 2.4 Device Global Memory and Data Transfer
- 2.5 Kernel Functions and Threading
- 2.6 Kernel Launch
- 2.7 Summary
- References
Chapter 3. Scalable parallel execution
- Abstract
- 3.1 CUDA Thread Organization
- 3.2 Mapping Threads to Multidimensional Data
- 3.3 Image Blur: A More Complex Kernel
- 3.4 Synchronization and Transparent Scalability
- 3.5 Resource Assignment
- 3.6 Querying Device Properties
- 3.7 Thread Scheduling and Latency Tolerance
- 3.8 Summary
Chapter 4. Memory and data locality
- Abstract
- 4.1 Importance of Memory Access Efficiency
- 4.2 Matrix Multiplication
- 4.3 CUDA Memory Types
- 4.4 Tiling for Reduced Memory Traffic
- 4.5 A Tiled Matrix Multiplication Kernel
- 4.6 Boundary Checks
- 4.7 Memory as a Limiting Factor to Parallelism
- 4.8 Summary
Chapter 5. Performance considerations
- Abstract
- 5.1 Global Memory Bandwidth
- 5.2 More on Memory Parallelism
- 5.3 Warps and SIMD Hardware
- 5.4 Dynamic Partitioning of Resources
- 5.5 Thread Granularity
- 5.6 Summary
- References
Chapter 6. Numerical considerations
- Abstract
- 6.1 Floating-Point Data Representation
- 6.2 Representable Numbers
- 6.3 Special Bit Patterns and Precision in IEEE Format
- 6.4 Arithmetic Accuracy and Rounding
- 6.5 Algorithm Considerations
- 6.6 Linear Solvers and Numerical Stability
- 6.7 Summary
- References
Chapter 7. Parallel patterns: convolution: An introduction to stencil computation
- Abstract
- 7.1 Background
- 7.2 1D Parallel Convolution—A Basic Algorithm
- 7.3 Constant Memory and Caching
- 7.4 Tiled 1D Convolution with Halo Cells
- 7.5 A Simpler Tiled 1D Convolution—General Caching
- 7.6 Tiled 2D Convolution With Halo Cells
- 7.7 Summary
- 7.8 Exercises
Chapter 8. Parallel patterns: prefix sum: An introduction to work efficiency in parallel algorithms
- Abstract
- 8.1 Background
- 8.2 A Simple Parallel Scan
- 8.3 Speed and Work Efficiency
- 8.4 A More Work-Efficient Parallel Scan
- 8.5 An Even More Work-Efficient Parallel Scan
- 8.6 Hierarchical Parallel Scan for Arbitrary-Length Inputs
- 8.7 Single-Pass Scan for Memory Access Efficiency
- 8.8 Summary
- 8.9 Exercises
- References
Chapter 9. Parallel patterns—parallel histogram computation: An introduction to atomic operations and privatization
- Abstract
- 9.1 Background
- 9.2 Use of Atomic Operations
- 9.3 Block versus Interleaved Partitioning
- 9.4 Latency versus Throughput of Atomic Operations
- 9.5 Atomic Operation in Cache Memory
- 9.6 Privatization
- 9.7 Aggregation
- 9.8 Summary
- Reference
Chapter 10. Parallel patterns: sparse matrix computation: An introduction to data compression and regularization
- Abstract
- 10.1 Background
- 10.2 Parallel SpMV Using CSR
- 10.3 Padding and Transposition
- 10.4 Using a Hybrid Approach to Regulate Padding
- 10.5 Sorting and Partitioning for Regularization
- 10.6 Summary
- References
Chapter 11. Parallel patterns: merge sort: An introduction to tiling with dynamic input data identification
- Abstract
- 11.1 Background
- 11.2 A Sequential Merge Algorithm
- 11.3 A Parallelization Approach
- 11.4 Co-Rank Function Implementation
- 11.5 A Basic Parallel Merge Kernel
- 11.6 A Tiled Merge Kernel
- 11.7 A Circular-Buffer Merge Kernel
- 11.8 Summary
- Reference
Chapter 12. Parallel patterns: graph search
- Abstract
- 12.1 Background
- 12.2 Breadth-First Search
- 12.3 A Sequential BFS Function
- 12.4 A Parallel BFS Function
- 12.5 Optimizations
- 12.6 Summary
- References
Chapter 13. CUDA dynamic parallelism
- Abstract
- 13.1 Background
- 13.2 Dynamic Parallelism Overview
- 13.3 A Simple Example
- 13.4 Memory Data Visibility
- 13.5 Configurations and Memory Management
- 13.6 Synchronization, Streams, and Events
- 13.7 A More Complex Example
- 13.8 A Recursive Example
- 13.9 Summary
- References
- A13.1 Code Appendix
Chapter 14. Application case study—non-Cartesian magnetic resonance imaging: An introduction to statistical estimation methods
- Abstract
- 14.1 Background
- 14.2 Iterative Reconstruction
- 14.3 Computing F^HD
- 14.4 Final Evaluation
- References
Chapter 15. Application case study—molecular visualization and analysis
- Abstract
- 15.1 Background
- 15.2 A Simple Kernel Implementation
- 15.3 Thread Granularity Adjustment
- 15.4 Memory Coalescing
- 15.5 Summary
- References
Chapter 16. Application case study—machine learning
- Abstract
- 16.1 Background
- 16.2 Convolutional Neural Networks
- 16.3 Convolutional Layer: A Basic CUDA Implementation of Forward Propagation
- 16.4 Reduction of Convolutional Layer to Matrix Multiplication
- 16.5 cuDNN Library
- References
Chapter 17. Parallel programming and computational thinking
- Abstract
- 17.1 Goals of Parallel Computing
- 17.2 Problem Decomposition
- 17.3 Algorithm Selection
- 17.4 Computational Thinking
- 17.5 Single Program, Multiple Data, Shared Memory and Locality
- 17.6 Strategies for Computational Thinking
- 17.7 A Hypothetical Example: Sodium Map of the Brain
- 17.8 Summary
- References
Chapter 18. Programming a heterogeneous computing cluster
- Abstract
- 18.1 Background
- 18.2 A Running Example
- 18.3 Message Passing Interface Basics
- 18.4 Message Passing Interface Point-to-Point Communication
- 18.5 Overlapping Computation and Communication
- 18.6 Message Passing Interface Collective Communication
- 18.7 CUDA-Aware Message Passing Interface
- 18.8 Summary
- Reference
Chapter 19. Parallel programming with OpenACC
- Abstract
- 19.1 The OpenACC Execution Model
- 19.2 OpenACC Directive Format
- 19.3 OpenACC by Example
- 19.4 Comparing OpenACC and CUDA
- 19.5 Interoperability with CUDA and Libraries
- 19.6 The Future of OpenACC
Chapter 20. More on CUDA and graphics processing unit computing
- Abstract
- 20.1 Model of Host/Device Interaction
- 20.2 Kernel Execution Control
- 20.3 Memory Bandwidth and Compute Throughput
- 20.4 Programming Environment
- 20.5 Future Outlook
- References
Chapter 21. Conclusion and outlook
- Abstract
- 21.1 Goals Revisited
- 21.2 Future Outlook
Appendix A. An introduction to OpenCL
- A.1 Background
- A.2 Data Parallelism Model
- A.3 Device Architecture
- A.4 Kernel Functions
- A.5 Device Management and Kernel Launch
- A.6 Electrostatic Potential Map in OpenCL
- A.7 Summary
Appendix B. THRUST: a productivity-oriented library for CUDA
- B.1 Background
- B.2 Motivation
- B.3 Basic Thrust Features
- B.4 Generic Programming
- B.5 Benefits of Abstraction
- B.6 Best Practices
Appendix C. CUDA Fortran
- C.1 CUDA Fortran and CUDA C Differences
- C.2 A First CUDA Fortran Program
- C.3 Multidimensional Array in CUDA Fortran
- C.4 Overloading Host/Device Routines with Generic Interfaces
- C.5 Calling CUDA C via ISO_C_Binding
- C.6 Kernel Loop Directives and Reduction Operations
- C.7 Dynamic Shared Memory
- C.8 Asynchronous Data Transfers
- C.9 Compilation and Profiling
- C.10 Calling Thrust from CUDA Fortran
Appendix D. An introduction to C++ AMP
- D.1 Core C++ AMP Features
- D.2 Details of the C++ AMP Execution Model
- D.3 Managing Accelerators
- D.4 Tiled Execution
- D.5 C++ AMP Graphics Features
- D.6 Summary
- Reference
Index

David B. Kirk

David B. Kirk is well recognized for his contributions to graphics hardware and algorithm research. By the time he began his studies at Caltech, he had already earned B.S. and M.S. degrees in mechanical engineering from MIT and worked as an engineer for Raster Technologies and Hewlett-Packard's Apollo Systems Division, and after receiving his doctorate, he joined Crystal Dynamics, a video-game manufacturing company, as chief scientist and head of technology. In 1997, he took the position of Chief Scientist at NVIDIA, a leader in visual computing technologies, and he is currently an NVIDIA Fellow.

At NVIDIA, Kirk led graphics-technology development for some of today's most popular consumer-entertainment platforms, playing a key role in providing mass-market graphics capabilities previously available only on workstations costing hundreds of thousands of dollars. For his role in bringing high-performance graphics to personal computers, Kirk received the 2002 Computer Graphics Achievement Award from the Association for Computing Machinery and the Special Interest Group on Graphics and Interactive Technology (ACM SIGGRAPH) and, in 2006, was elected to the National Academy of Engineering, one of the highest professional distinctions for engineers.

Kirk holds 50 patents and patent applications relating to graphics design and has published more than 50 articles on graphics technology, won several best-paper awards, and edited the book Graphics Gems III. A technological "evangelist" who cares deeply about education, he has supported new curriculum initiatives at Caltech and has been a frequent university lecturer and conference keynote speaker worldwide.

Affiliations and expertise

NVIDIA Fellow

Wen-mei W. Hwu

Wen-mei W. Hwu is a Professor and holds the Sanders-AMD Endowed Chair in the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign. His research interests are in the area of architecture, implementation, compilation, and algorithms for parallel computing. He is the chief scientist of Parallel Computing Institute and director of the IMPACT research group (www.impact.crhc.illinois.edu). He is a co-founder and CTO of MulticoreWare. For his contributions in research and teaching, he received the ACM SigArch Maurice Wilkes Award, the ACM Grace Murray Hopper Award, the Tau Beta Pi Daniel C. Drucker Eminent Faculty Award, the ISCA Influential Paper Award, the IEEE Computer Society B. R. Rau Award and the Distinguished Alumni Award in Computer Science of the University of California, Berkeley. He is a fellow of IEEE and ACM. He directs the UIUC CUDA Center of Excellence and serves as one of the principal investigators of the NSF Blue Waters Petascale computer project. Dr. Hwu received his Ph.D. degree in Computer Science from the University of California, Berkeley.

Affiliations and expertise

CTO, MulticoreWare and professor specializing in compiler design, computer architecture, microarchitecture, and parallel processing, University of Illinois at Urbana-Champaign, USA

Life Sciences

Physical Sciences & Engineering

Social Sciences & Humanities

Health