GPU Programming with OpenCL Training Course

Introduction

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

Getting Started

• Initialising a new OpenCL project in Visual Studio Code
• Examining project structure and file organisation
• Compiling and executing the programme
• Outputting results using printf and fprintf

OpenCL API

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

OpenCL C

• The role of OpenCL C in device-side programming
• Writing kernels in OpenCL C to execute on the device and manipulate data
• Utilising OpenCL C data types, qualifiers, operators, and expressions
• Applying built-in functions for mathematical, geometric, and relational operations
• Working with extensions and libraries such as atomic operations, image processing, and cl_khr_fp16

OpenCL Memory Model

• Distinguishing between host and device memory models
• Working with memory spaces including global, local, constant, and private
• Managing memory objects such as buffers, images, and pipes
• Configuring memory access modes like read-only, write-only, and read-write
• Implementing memory consistency models and synchronisation mechanisms

OpenCL Execution Model

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

Debugging

• Identifying common errors and bugs in OpenCL programmes
• Using the Visual Studio Code debugger to inspect variables, breakpoints, and the call stack
• Debugging and analysing OpenCL programmes on AMD devices with CodeXL
• Debugging and analysing OpenCL programmes on Intel devices with Intel VTune
• Debugging and analysing OpenCL programmes on NVIDIA devices with NVIDIA Nsight

Optimisation

• Understanding factors that influence OpenCL programme performance
• Improving arithmetic throughput with vector data types and vectorisation techniques
• Reducing control overhead and increasing locality through loop unrolling and tiling
• Optimising memory access and bandwidth using local memory and related functions
• Measuring and enhancing execution time and resource utilisation via OpenCL profiling tools