GPU Programming with OpenCL Training Course

Introduction

• Overview of OpenCL
• Comparing OpenCL, CUDA, and SYCL
• Key features and architectural components of OpenCL
• Configuring the development environment

Getting Started

• Creating a new OpenCL project in Visual Studio Code
• Navigating project structure and essential files
• Compiling and executing the program
• Displaying output using printf and fprintf functions

OpenCL API

• The role of the OpenCL API within host programs
• Querying device information and capabilities via the API
• Creating contexts, command queues, buffers, kernels, and events
• Enqueueing commands such as read, write, copy, map, unmap, execute, and wait
• Handling errors and exceptions effectively

OpenCL C

• The role of OpenCL C in device-side programming
• Writing kernels for device execution and data manipulation
• Utilizing data types, qualifiers, operators, and expressions in OpenCL C
• Employing built-in functions for math, geometry, and relational operations
• Using extensions and libraries such as atomic, image, and cl_khr_fp16

OpenCL Memory Model

• Distinguishing between host and device memory models
• Working with memory spaces: global, local, constant, and private
• Managing memory objects like buffers, images, and pipes
• Applying memory access modes: read-only, write-only, and read-write
• Understanding the memory consistency model and synchronization mechanisms

OpenCL Execution Model

• Differences between host and device execution models
• Defining parallelism using work-items, work-groups, and ND-ranges
• Using work-item functions like get_global_id, get_local_id, and get_group_id
• Utilizing work-group functions such as barrier, work_group_reduce, and work_group_scan
• Applying device functions including get_num_groups, get_global_size, and get_local_size

Debugging

• Identifying common errors and bugs in OpenCL applications
• Inspecting variables, breakpoints, and call stacks using the Visual Studio Code debugger
• Debugging and analyzing OpenCL programs on AMD devices with CodeXL
• Debugging and analyzing OpenCL programs on Intel devices using Intel VTune
• Debugging and analyzing OpenCL programs on NVIDIA devices with NVIDIA Nsight

Optimization

• Factors influencing OpenCL program performance
• Improving arithmetic throughput using vector data types and vectorization
• Reducing control overhead and increasing locality via loop unrolling and tiling
• Optimizing memory access and bandwidth with local memory and related functions
• Measuring and improving execution time and resource utilization through profiling tools