Introduction to GPU Programming Training Course

Introduction

• Defining GPU programming
• Rationale for utilizing GPU programming
• Challenges and trade-offs associated with GPU development
• Overview of available frameworks and tools for GPU programming
• Selecting the optimal framework and tool for specific applications

OpenCL

• Introduction to OpenCL
• Advantages and limitations of OpenCL
• Configuring the OpenCL development environment
• Developing a basic OpenCL program for vector addition
• Using the OpenCL API to query device details, manage memory, transfer data, launch kernels, and synchronize threads
• Writing device-executed kernels using the OpenCL C language to manipulate data
• Utilizing OpenCL built-in functions, variables, and libraries for common operations
• Optimizing data transfers and memory access using OpenCL memory spaces (global, local, constant, and private)
• Controlling parallelism via the OpenCL execution model (work-items, work-groups, and ND-ranges)
• Debugging and testing OpenCL programs with tools such as CodeXL
• Enhancing OpenCL program performance through coalescing, caching, prefetching, and profiling

CUDA

• Introduction to CUDA
• Advantages and limitations of CUDA
• Setting up the CUDA development environment
• Creating a basic CUDA program for vector addition
• Using the CUDA API to query device information, manage memory, copy data, launch kernels, and synchronize threads
• Writing device-executed kernels using the CUDA C/C++ language to manipulate data
• Leveraging CUDA built-in functions, variables, and libraries for standard tasks
• Optimizing data flows using CUDA memory spaces (global, shared, constant, and local)
• Managing parallelism through the CUDA execution model (threads, blocks, and grids)
• Debugging and testing CUDA programs using CUDA-GDB, CUDA-MEMCHECK, and NVIDIA Nsight
• Optimizing CUDA program performance via coalescing, caching, prefetching, and profiling

ROCm

• Introduction to ROCm
• Advantages and limitations of ROCm
• Configuring the ROCm development environment
• Developing a basic ROCm program for vector addition
• Using the ROCm API to query device information, manage memory, transfer data, launch kernels, and synchronize threads
• Writing device-executed kernels using the ROCm C/C++ language to manipulate data
• Utilizing ROCm built-in functions, variables, and libraries for common operations
• Optimizing data transfers using ROCm memory spaces (global, local, constant, and private)
• Controlling parallelism via the ROCm execution model (threads, blocks, and grids)
• Debugging and testing ROCm programs using the ROCm Debugger and ROCm Profiler
• Enhancing ROCm program performance through coalescing, caching, prefetching, and profiling

HIP

• Introduction to HIP
• Advantages and limitations of HIP
• Setting up the HIP development environment
• Creating a basic HIP program for vector addition
• Writing device-executed kernels using the HIP language to manipulate data
• Leveraging HIP built-in functions, variables, and libraries for standard tasks
• Optimizing data flows using HIP memory spaces (global, shared, constant, and local)
• Managing parallelism through the HIP execution model (threads, blocks, and grids)
• Debugging and testing HIP programs using the ROCm Debugger and ROCm Profiler
• Optimizing HIP program performance via coalescing, caching, prefetching, and profiling

Comparison

• Comparing the features, performance characteristics, and compatibility of OpenCL, CUDA, ROCm, and HIP
• Evaluating GPU programs through benchmarks and performance metrics
• Learning best practices and strategic tips for effective GPU programming
• Exploring current trends and future challenges in GPU programming

Summary and Next Steps