CUDA Programming Course Practicals

Course materials from the CUDA Programming on NVIDIA GPUs course taught by Prof. Mike Giles and Prof. Wes Armour at the University of Oxford.

Course Website: https://people.maths.ox.ac.uk/gilesm/cuda/ Capstone Project: https://github.com/MaxMLang/cuda-gis-smoothing/

Overview

This repository contains my work on the practical exercises from the CUDA programming course. The code includes:

• Original examples from the course
• My own implementations and optimizations
• Modified versions with performance improvements
• A capstone project applying multiple concepts

Practical Structure

Core Practicals (1-12)

• prac1: Hello world examples - kernel launching, data transfer, error checking
• prac2: Monte Carlo simulation using cuRAND - constant memory, random generation, bandwidth optimization
• prac3: 3D Laplace solver - thread block optimization, memory layout, profiling
• prac4: Reduction operations - shared memory, synchronization, shuffles, atomics
• prac5: Tensor Cores and cuBLAS - library usage and optimization
• prac6: C++ integration - templates, libraries, mixed compilation
• prac7: Tri-diagonal equations - specialized algorithms
• prac8: Scan operations - recurrence equations and parallel algorithms
• prac9: Pattern matching - string processing on GPU
• prac10: Auto-tuning - performance optimization techniques
• prac11: Streams and OpenMP - asynchronous processing and multithreading
• prac12: Advanced streaming - computation/communication overlap

Key Learning Areas

• Memory management (global, shared, constant, texture)
• Thread organization and synchronization
• Performance optimization and profiling
• CUDA libraries (cuBLAS, cuRAND, cuFFT)
• Multi-GPU programming
• Debugging and error handling

Course Prerequisites

• C/C++ programming experience
• No prior parallel computing knowledge required
• Access to NVIDIA GPU hardware

Building and Running

Each practical directory contains:

• Source files (.cu, .cpp)
• Makefile for compilation
• Shell script for execution
• Answer files documenting solutions and insights

Capstone Project

A satellite data smoothing case study that integrates multiple concepts from the practicals, demonstrating real-world application of CUDA programming techniques.

Notes

• Some implementations are direct from course materials
• Others include my own optimizations and modifications
• Performance improvements and alternative approaches documented in answer files
• All work completed as part of the course curriculum