AMD has just released an update to their AMD programming guide.
Download the guide (PDF) August version
Download the guide (PDF) November version
For more optimisation guides, see the tutorials page of the knowledge base.
Table of Contents
Chapter 1 OpenCL Architecture and AMD Accelerated Parallel Processing
1.1 Software Overview
1.1.1 Synchronization
1.2 Hardware Overview for Southern Islands Devices
1.3 Hardware Overview for Evergreen and Northern Islands Devices
1.4 The AMD Accelerated Parallel Processing Implementation of OpenCL
1.4.1 Work-Item Processing
1.4.2 Flow Control
1.4.3 Work-Item Creation
1.5 Memory Architecture and Access
1.5.1 Memory Access
1.5.2 Global Buffer
1.5.3 Image Read/Write
1.5.4 Memory Load/Store
1.6 Communication Between Host and
1.6.1 PCI Express Bus
1.6.2 Processing API Calls: The Command Processor
1.6.3 DMA Transfers
1.6.4 Masking Visible Devices
1.7 GPU Compute Device Scheduling
1.8 Terminology
1.8.1 Compute Kernel
1.8.2 Wavefronts and Work-groups
1.8.3 Local Data Store (LDS)
1.9 Programming Model
1.10 Example Programs
1.10.1 First Example: Simple Buffer Write
1.10.2 Example: Parallel Min() Function .1-23
Chapter 2 Building and Running OpenCL Programs
2.1 Compiling the Program
2.1.1 Compiling on Windows
2.1.2 Compiling on Linux
2.1.3 Supported Standard OpenCL Compiler Options
2.1.4 AMD-Developed Supplemental Compiler Options
2.2 Running the Program
2.2.1 Running Code on Windows
2.2.2 Running Code on Linux
2.3 Calling Conventions
Chapter 3 Debugging OpenCL
3.1 AMD CodeXL GPU Debugger .
3.2 Debugging CPU Kernels with GDB
3.2.1 Setting the Environment .3-
3.2.2 Setting the Breakpoint in an OpenCL Kernel.3-2
3.2.3 Sample GDB Session
3.2.4 Notes
Chapter 4 OpenCL Performance and Optimization
4.1 CodeXL GPU Profiler
4.1.1 Collecting OpenCL Application Traces
4.1.2 Timeline View.
4.1.3 Summary Pages View
4.1.4 API Trace View
4.1.5 Collecting OpenCL GPU Kernel Performance Counters
4.2 AMD APP KernelAnalyzer2
4.2.1 Start KernelAnalyzer2
4.2.2 Open Kernel Source
4.2.3 Build Options – Choosing Target ASICS
4.2.4 Build Options – Defining Kernel Compilation Options
4.2.5 Analysis Input Tab
4.2.6 Build the Kernel
4.2.7 Build Statistics Tab
4.2.8 The Analysis Tab
4.3 Analyzing Processor Kernels
4.3.1 Intermediate Language and GPU Disassembly
4.3.2 Generating IL and ISA Code.
4.4 Estimating Performance
4.4.1 Measuring Execution Time
4.4.2 Using the OpenCL timer with Other System Timers
4.4.3 Estimating Memory Bandwidth
4.5 OpenCL Memory Objects
4.5.1 Types of Memory Used by the Runtime
4.5.2 Placement
4.5.3 Memory Allocation
4.5.4 Mapping
4.5.5 Reading, Writing, and Copying
4.5.6 Command Queue
4.6 OpenCL Data Transfer Optimization
4.6.1 Definitions
4.6.2 Buffers
4.7 Using Multiple OpenCL Devices
4.7.1 CPU and GPU Devices
4.7.2 When to Use Multiple Devices
4.7.3 Partitioning Work for Multiple Devices
4.7.4 Synchronization Caveats
4.7.5 GPU and CPU Kernels
4.7.6 Contexts and Devices
Chapter 5 OpenCL Performance and Optimization for GCN Devices
5.1 Global Memory Optimization
5.1.1 Channel Conflicts.
5.1.2 Coalesced Writes
5.1.3 Hardware Variations.
5.2 Local Memory (LDS) Optimization
5.3 Constant Memory Optimization
5.4 OpenCL Memory Resources: Capacity and Performance
5.5 Using LDS or L1 Cache
5.6 NDRange and Execution Range Optimization
5.6.1 Hiding ALU and Memory Latency
5.6.2 Resource Limits on Active Wavefronts
5.6.3 Partitioning the Work
5.6.4 Summary of NDRange Optimizations
5.7 Instruction Selection Optimizations
5.7.1 Instruction Bandwidths
5.7.2 AMD Media Instructions
5.7.3 Math Libraries
5.7.4 Compiler Optimizations
5.8 Additional Performance Guidance
5.8.1 Loop Unroll pragma
5.8.2 Memory Tiling
5.8.3 General Tips
5.8.4 Guidance for CUDA Programmers Using OpenCL
5.8.5 Guidance for CPU Programmers Using OpenCL to Program GPUs
5.8.6 Optimizing Kernel Code
5.8.7 Optimizing Kernels for Southern Island GPUs
5.9 Specific Guidelines for Southern Islands GPUs
Chapter 6 OpenCL Performance and Optimization for Evergreen and Northern Islands Devices
6.1 Global Memory Optimization
6.1.1 Two Memory Paths
6.1.2 Channel Conflicts.
6.1.3 Float4 Or Float1
6.1.4 Coalesced Writes
6.1.5 Alignment
6.1.6 Summary of Copy Performance
6.1.7 Hardware Variations
6.2 Local Memory (LDS) Optimization
6.3 Constant Memory Optimization
6.4 OpenCL Memory Resources: Capacity and Performance
6.5 Using LDS or L1 Cache
6.6 NDRange and Execution Range Optimization
6.6.1 Hiding ALU and Memory Latency
6.6.2 Resource Limits on Active Wavefronts
6.6.3 Partitioning the Work
6.6.4 Optimizing for Cedar
6.6.5 Summary of NDRange Optimizations
6.7 Using Multiple OpenCL Devices
6.7.1 CPU and GPU Devices
6.7.2 When to Use Multiple Devices
6.7.3 Partitioning Work for Multiple Devices
6.7.4 Synchronization Caveats
6.7.5 GPU and CPU Kernels
6.7.6 Contexts and Devices
6.8 Instruction Selection Optimizations
6.8.1 Instruction Bandwidths
6.8.2 AMD Media Instructions
6.8.3 Math Libraries
6.8.4 VLIW and SSE Packing
6.8.5 Compiler Optimizations
6.9 Clause Boundaries
6.10 Additional Performance Guidance
6.10.1 Loop Unroll pragma
6.10.2 Memory Tiling
6.10.3 General Tips.
6.10.4 Guidance for CUDA Programmers Using OpenCL
6.10.5 Guidance for CPU Programmers Using OpenCL to Program GPUs
6.10.6 Optimizing Kernel Code
6.10.7 Optimizing Kernels for Evergreen and 69XX-Series GPUs
Chapter 7 OpenCL Static C++ Programming Language
7.1 Overview
7.1.1 Supported Features
7.1.2 Unsupported Features
7.1.3 Relations with ISO/IEC C++
7.2 Additions and Changes to Section 5 – The OpenCL C Runtime
7.2.1 Additions and Changes to Section 5.
7.1 – Creating Kernel Objects
7.2.2 Passing Classes between Host and Device
7.3 Additions and Changes to Section The OpenCL C Programming Language
7.3.1 Building C++ Kernels
7.3.2 Classes and Derived Classes
7.3.3 Namespaces
7.3.4 Overloading
7.3.5 Templates
7.3.6 Exceptions
7.3.7 Libraries
7.3.8 Dynamic Operation
7.4 Examples
7.4.1 Passing a Class from the Host to the Device and Back
7.4.2 Kernel Overloading
7.4.3 Kernel Template
Appendix A OpenCL Optional Extensions
A.1 Extension Name Convention
A.2 Querying Extensions for a Platform
A.3 Querying Extensions for a Device
A.4 Using Extensions in Kernel Programs
A.5 Getting Extension Function Pointers
A.6 List of Supported Extensions that are Khronos-Approved
A.7 cl_ext Extensions
A.8 AMD Vendor-Specific Extensions
A.8.1 cl_amd_fp64
A.8.2 cl_amd_vec3
A.8.3 cl_amd_device_persistent_memory
A.8.4 cl_amd_device_attribute_query
A.8.5 cl_amd_compile_options
A.8.6 cl_amd_offline_devices
A.8.7 cl_amd_event_callback
A.8.8 cl_amd_popcnt
A.8.9 cl_amd_media_ops
A.8.10 cl_amd_media_ops2
A.8.11 cl_amd_printf
A.8.12 cl_amd_predefined_macros
A.8.13 cl_amd_bus_addressable_memory
A.9 Supported Functions for cl_amd_fp64/cl_khr_fp64
A.10 Extension Support by Device
Appendix B The OpenCL Installable Client Driver (ICD)
B.1 Overview
B.2 Using ICD
Appendix C Compute Kernel
C.1 Differences from a Pixel Shader
C.2 Indexing
C.3 Performance Comparison
C.4 Pixel Shader
C.5 Compute Kernel
C.6 LDS Matrix Transpose
C.7 Results Comparison
Appendix D Device Parameters
Appendix E OpenCL Binary Image Format (BIF) v2.0
E.1 Overview
E.1.1 Executable and Linkable Format (ELF) Header
E.1.2 Bitness
E.2 BIF Options
Appendix F Open Decode API Tutorial
F.1 Overview
F.2 Initializing
F.3 Creating the Context
F.4 Creating the Session
F.5 Decoding
F.6 Destroying Session and Context
Appendix G OpenCL-OpenGL Interoperability
G.1 Under Windows
G.1.1 Single GPU Environment
G.1.2 Multi-GPU Environment
G.1.3 Limitations
G.2 Linux Operating System
G.2.1 Single GPU Environment
G.2.2 Multi-GPU Configuration
G.3 Additional GL Formats Supported
