CUDA Application Design and Development.

Mostrar otras versiones (3)

As the computer industry retools to leverage massively parallel graphics processing units (GPUs), this book is designed to meet the needs of working software developers who need to understand GPU programming with CUDA and increase efficiency in their projects. CUDA Application Design and Development...

Descripción completa

Detalles Bibliográficos
Clasificación:Libro Electrónico
Autor principal: Farber, Rob
Formato: Electrónico eBook
Idioma:Inglés
Publicado: Burlington : Elsevier Science, 2011.
Temas:
Application design.
Application software > Development.
Application software.
Computer architecture.
Parallel programming (Computer science)
Logiciels d'application > Développement.
Logiciels d'application.
Ordinateurs > Architecture.
Programmation parallèle (Informatique)
Application software
Application software > Development
Computer architecture
Acceso en línea:Texto completo
Tabla de Contenidos:
  • Front Cover
  • CUDA Application Design and Development
  • Copyright
  • Dedication
  • Table of Contents
  • Foreword
  • Preface
  • 1 First Programs and How to Think in CUDA
  • Source Code and Wiki
  • Distinguishing CUDA from Conventional Programming with a Simple Example
  • Choosing a CUDA API
  • Some Basic CUDA Concepts
  • Understanding Our First Runtime Kernel
  • Three Rules of GPGPU Programming
  • Rule 1: Get the Data on the GPU and Keep It There
  • Rule 2: Give the GPGPU Enough Work to Do
  • Rule 3: Focus on Data Reuse within the GPGPU to Avoid Memory Bandwidth Limitations
  • Big-O Considerations and Data Transfers
  • CUDA and Amdahl's Law
  • Data and Task Parallelism
  • Hybrid Execution: Using Both CPU and GPU Resources
  • Regression Testing and Accuracy
  • Silent Errors
  • Introduction to Debugging
  • UNIX Debugging
  • NVIDIA's cuda-gdb Debugger
  • The CUDA Memory Checker
  • Use cuda-gdb with the UNIX ddd Interface
  • Windows Debugging with Parallel Nsight
  • Summary
  • 2 CUDA for Machine Learning and Optimization
  • Modeling and Simulation
  • Fitting Parameterized Models
  • Nelder-Mead Method
  • Levenberg-Marquardt Method
  • Algorithmic Speedups
  • Machine Learning and Neural Networks
  • XOR: An Important Nonlinear Machine-Learning Problem
  • An Example Objective Function
  • A Complete Functor for Multiple GPU Devices and the Host Processors
  • Brief Discussion of a Complete Nelder-Mead Optimization Code
  • Performance Results on XOR
  • Performance Discussion
  • Summary
  • The C++ Nelder-Mead Template
  • 3 The CUDA Tool Suite: Profiling a PCA/NLPCA Functor
  • PCA and NLPCA
  • Autoencoders
  • An Example Functor for PCA Analysis
  • An Example Functor for NLPCA Analysis
  • Obtaining Basic Profile Information
  • Gprof: A Common UNIX Profiler
  • The NVIDIA Visual Profiler: Computeprof
  • Parallel Nsight for Microsoft Visual Studio.
  • The Nsight Timeline Analysis
  • The NVTX Tracing Library
  • Scaling Behavior of the CUDA API
  • Tuning and Analysis Utilities (TAU)
  • Summary
  • 4 The CUDA Execution Model
  • GPU Architecture Overview
  • Thread Scheduling: Orchestrating Performance and Parallelism via the Execution Configuration
  • Relevant computeprof Values for a Warp
  • Warp Divergence
  • Guidelines for Warp Divergence
  • Relevant computeprof Values for Warp Divergence
  • Warp Scheduling and TLP
  • Relevant computeprof Values for Occupancy
  • ILP: Higher Performance at Lower Occupancy
  • ILP Hides Arithmetic Latency
  • ILP Hides Data Latency
  • ILP in the Future
  • Relevant computeprof Values for Instruction Rates
  • Little's Law
  • CUDA Tools to Identify Limiting Factors
  • The nvcc Compiler
  • Launch Bounds
  • The Disassembler
  • PTX Kernels
  • GPU Emulators
  • Summary
  • 5 CUDA Memory
  • The CUDA Memory Hierarchy
  • GPU Memory
  • L2 Cache
  • Relevant computeprof Values for the L2 Cache
  • L1 Cache
  • Relevant computeprof Values for the L1 Cache
  • CUDA Memory Types
  • Registers
  • Local memory
  • Relevant computeprof Values for Local Memory Cache
  • Shared Memory
  • Relevant computeprof Values for Shared Memory
  • Constant Memory
  • Texture Memory
  • Relevant computeprof Values for Texture Memory
  • Global Memory
  • Common Coalescing Use Cases
  • Allocation of Global Memory
  • Limiting Factors in the Design of Global Memory
  • Relevant computeprof Values for Global Memory
  • Summary
  • 6 Efficiently Using GPU Memory
  • Reduction
  • The Reduction Template
  • A Test Program for functionReduce.h
  • Results
  • Utilizing Irregular Data Structures
  • Sparse Matrices and the CUSP Library
  • Graph Algorithms
  • SoA, AoS, and Other Structures
  • Tiles and Stencils
  • Summary
  • 7 Techniques to Increase Parallelism
  • CUDA Contexts Extend Parallelism
  • Streams and Contexts.
  • Multiple GPUs
  • Explicit Synchronization
  • Implicit Synchronization
  • The Unified Virtual Address Space
  • A Simple Example
  • Profiling Results
  • Out-of-Order Execution with Multiple Streams
  • Tip for Concurrent Kernel Execution on the Same GPU
  • Atomic Operations for Implicitly Concurrent Kernels
  • Tying Data to Computation
  • Manually Partitioning Data
  • Mapped Memory
  • How Mapped Memory Works
  • Summary
  • 8 CUDA for All GPU and CPU Applications
  • Pathways from CUDA to Multiple Hardware Backends
  • The PGI CUDA x86 Compiler
  • The PGI CUDA x86 Compiler
  • An x86 core as an SM
  • The NVIDIA NVCC Compiler
  • Ocelot
  • Swan
  • MCUDA
  • Accessing CUDA from Other Languages
  • SWIG
  • Copperhead
  • EXCEL
  • MATLAB
  • Libraries
  • CUBLAS
  • CUFFT
  • MAGMA
  • phiGEMM Library
  • CURAND
  • Summary
  • 9 Mixing CUDA and Rendering
  • OpenGL
  • GLUT
  • Mapping GPU Memory with OpenGL
  • Using Primitive Restart for 3D Performance
  • Introduction to the Files in the Framework
  • The Demo and Perlin Example Kernels
  • The Demo Kernel
  • The Demo Kernel to Generate a Colored Sinusoidal Surface
  • Perlin Noise
  • Using the Perlin Noise Kernel to Generate Artificial Terrain
  • The simpleGLmain.cpp File
  • The simpleVBO.cpp File
  • The callbacksVBO.cpp File
  • Summary
  • 10 CUDA in a Cloud and Cluster Environments
  • The Message Passing Interface (MPI)
  • The MPI Programming Model
  • The MPI Communicator
  • MPI Rank
  • Master-Slave
  • Point-to-Point Basics
  • How MPI Communicates
  • Bandwidth
  • Balance Ratios
  • Considerations for Large MPI Runs
  • Scalability of the Initial Data Load
  • Using MPI to Perform a Calculation
  • Check Scalability
  • Cloud Computing
  • A Code Example
  • Data Generation
  • Summary
  • 11 CUDA for Real Problems
  • Working with High-Dimensional Data
  • PCA/NLPCA
  • Multidimensional Scaling
  • K-Means Clustering.
  • Expectation-Maximization
  • Support Vector Machines
  • Bayesian Networks
  • Mutual information
  • Force-Directed Graphs
  • Monte Carlo Methods
  • Molecular Modeling
  • Quantum Chemistry
  • Interactive Workflows
  • A Plethora of Projects
  • Summary
  • 12 Application Focus on Live Streaming Video
  • Topics in Machine Vision
  • 3D Effects
  • Segmentation of Flesh-colored Regions
  • Edge Detection
  • FFmpeg
  • TCP Server
  • Live Stream Application
  • kernelWave(): An Animated Kernel
  • kernelFlat(): Render the Image on a Flat Surface
  • kernelSkin(): Keep Only Flesh-colored Regions
  • kernelSobel(): A Simple Sobel Edge Detection Filter
  • The launch_kernel() Method
  • The simpleVBO.cpp File
  • The callbacksVBO.cpp File
  • Building and Running the Code
  • The Future
  • Machine Learning
  • The Connectome
  • Summary
  • Listing for simpleVBO.cpp
  • Works Cited
  • Index
  • A
  • B
  • C
  • D
  • E
  • F
  • G
  • H
  • I
  • J
  • K
  • L
  • M
  • N
  • O
  • P
  • Q
  • R
  • S
  • T
  • U
  • V
  • W
  • X.

Ejemplares similares