Front Cover -- HeterogeneousComputing withOpenCL -- Copyright -- Contents -- Foreword to the RevisedOpenCL 1.2 Edition -- Foreword to the First Edition -- Preface -- Our Heterogeneous World -- OpenCL -- This Text -- Acknowledgments -- About the Authors -- Chapter 1: Introduction to Parallel Programming -- Introduction -- OpenCL -- The Goals of This Book -- Thinking Parallel -- Concurrency and Parallel Programming Models -- Threads and Shared Memory -- Message-Passing Communication -- Different Grains of Parallelism -- Data Sharing and Synchronization -- Structure -- Reference -- Further Reading and Relevant Websites -- Chapter 2: Introduction to OpenCL -- Introduction -- The OpenCL Standard -- The OpenCL Specification -- Kernels and the OpenCL Execution Model -- Platform and Devices -- Host-Device Interaction -- The Execution Environment -- Contexts -- Command Queues -- Events -- Memory Objects -- Buffers -- Images -- Flush and Finish -- Creating an OpenCL Program Object -- The OpenCL Kernel -- Memory Model -- Writing Kernels -- Full Source Code Example for Vector Addition -- Vector Addition with C++ Wrapper -- Summary -- Reference -- Chapter 3: OpenCL Device Architectures -- Introduction -- Hardware trade-offs -- Performance Increase by Frequency, and Its Limitations -- Superscalar Execution -- VLIW -- SIMD and Vector Processing -- Hardware Multithreading -- Multi-Core Architectures -- Integration: Systems-on-Chip and the APU -- Cache Hierarchies and Memory Systems -- The architectural design space -- CPU Designs -- Low-Power CPUs -- Mainstream Desktop CPUs -- Intel Itanium 2 -- Niagara -- GPU Architectures -- Handheld GPUs -- At the High End: AMD Radeon HD7970 and NVIDIA GTX580 -- APU and APU-Like Designs -- Summary -- References -- Chapter 4: Basic OpenCL Examples -- Introduction -- Example Applications -- Simple Matrix Multiplication Example.

Step 1: Set Up Environment -- Step 2: Declare Buffers and Move Data -- Step 3: Runtime Kernel Compilation -- Step 4: Run the Program -- Step 5: Return Results to Host -- Image Rotation Example -- Step 1: Set Up Environment -- Step 2: Declare Buffers and Move Data -- Step 3: Runtime Kernel Compilation -- Step 4: Run the Program -- Step 5: Read Result Back to Host -- Image Convolution Example -- Step 1: Create Image and Buffer Objects -- Step 2: Write the Input Data -- Step 3: Create Sampler Object -- Step 4: Compile and Execute the Kernel -- Step 5: Read the Result -- The Convolution Kernel -- Compiling OpenCL Host Applications -- Summary -- Chapter 5: Understanding OpenCL's Concurrency and Execution Model -- Introduction -- Kernels, Work-Items, Workgroups, and the Execution Domain -- OpenCL Synchronization: Kernels, Fences, and Barriers -- Queuing and Global Synchronization -- Memory Consistency in OpenCL -- Events -- Command Queues to Multiple Devices -- Event Uses beyond Synchronization -- User Events -- Event Callbacks -- Native Kernels -- Command Barriers and Markers -- The Host-Side Memory Model -- Buffers -- Manipulating Buffer Objects -- Images -- The Device-Side Memory Model -- Device-Side Relaxed Consistency -- Global Memory -- Local Memory -- Constant Memory -- Private Memory -- Summary -- Chapter 6: Dissecting a CPU/GPU OpenCL Implementation -- Introduction -- OpenCL on an AMD Bulldozer CPU -- OpenCL on the AMD Radeon HD7970 GPU -- Threading and the Memory System -- Instruction Execution on the HD7970 Architecture -- The Shift from VLIW Execution -- Resource Allocation -- Memory Performance Considerations in OpenCL -- OpenCL Global Memory -- Local Memory as a Software-Managed Cache -- Summary -- References -- Chapter 7: Data Management -- Memory management -- Data transfer in a discrete environment -- Optimizations -- Zero-Copy Buffers.

Data placement in a shared-memory environment -- Local Memory -- Cacheable System Memory -- Uncached System Memory -- Example application-work group reduction -- Using a Discrete GPU Device -- Case1 Using device buffers -- Case2 Using pinned staging buffers -- Case3 Using zero-copy buffers -- Case4 Combination -- Using an APU -- Case1 Using local memory buffers -- Case2 Using pinned staging buffers -- Case3 Using zero-copy buffers -- References -- Chapter 8: OpenCL Case Study -- Introduction -- Convolution Kernel -- Selecting Workgroup Sizes -- Caching Data to Local Memory -- Aligning for Memory Accesses -- Improving Efficiency with Vector Reads -- Performing the Convolution -- Improving Performance with Loop Unrolling -- Conclusions -- Code Listings -- Host Code -- Kernel Code -- Reference -- Chapter 9: OpenCL Case Study -- Introduction -- Choosing the Number of Workgroups -- Choosing the Optimal Workgroup Size -- Optimizing Global Memory Data Access Patterns -- Using Atomics to Perform Local Histogram -- Optimizing Local Memory Access -- Local Histogram Reduction -- The Global Reduction -- Full Kernel Code -- Performance and Summary -- Chapter 10: OpenCL Case Study -- Introduction -- Overview of the Computation -- GPU Implementation -- Buffer Creation -- Building the Acceleration Structure -- Computing Collisions -- Integration -- CPU Implementation -- Load Balancing -- Performance and Summary -- Kernel for Uniform Grid Creation -- Kernels for Simulation -- Chapter 11: OpenCL Extensions -- Introduction -- Overview of Extension Mechanism -- Device Fission -- Double Precision -- References -- Chapter 12: Foreign Lands -- Introduction -- Beyond C and C++ -- Haskell OpenCL -- Module Structure -- Environments -- Reference Counting -- Platform and Devices -- The Execution Environment -- Contexts -- Command Queues -- Buffers.

Creating an OpenCL Program Object -- The OpenCL Kernel -- Full Source Code Example for Vector Addition -- Summary -- References -- Chapter 13: OpenCL Profiling and Debugging -- Introduction -- Profiling with Events -- AMD Accelerated Parallel Processing Profiler -- Collecting OpenCL Application Trace -- Summary Pages View -- API Trace View -- Collecting OpenCL GPU Kernel Performance Counters -- AMD Accelerated Parallel Processing KernelAnalyzer -- Walking through the AMD APP Profiler -- Starting the AMD APP Profiler -- Using the Application Trace to Find the Application Bottleneck -- Using the GPU Performance Counters to Find the Bottleneck in the Kernel -- Debugging OpenCL Applications -- Overview of gDEBugger -- Debugging Parallel OpenCL Applications with gDEBugger -- API-Level Debugging -- Kernel Debugging -- AMD Printf Extension -- Conclusion -- Chapter Chapter 14: Performance Optimization of an Image Analysis Application -- Introduction -- Description of the algorithm -- Migrating multithreaded CPU implementation to OpenCL -- Hotspot Analysis -- Kernel Development and Static Analysis -- Performance optimization -- Kernel Occupancy -- Kernel Occupancy for AMD Radeon HD5000/6000 Series -- Kernel Occupancy for AMD Radeon HD 7000 -- Impact of Workgroup Size -- Impact of VGPR and LDS -- Power and performance analysis -- Conclusion -- References.