Description

This repository contains solutions for the university CUDA course. The solutions contain code samples with Cython + CUDA showing how to generate CUDA capable python extensions.

The repository is organized as follows:

• vector_addiction
  • A simple CUDA program that adds two vectors.
    • The most basic example of CUDA.
• matrix_mul (Lab2)
  • A simple CUBLAS program that multiplies two square matrices.
    • Uses CUBLAS for matrix multiplication.
• dotproduct
  • An implementation of dot product (or scalar product) in CUDA.
    • Uses shared memory for partial reduction.
• raytracing
  • A very simple implementation of raytracing with randomly generated spheres. No lighting, reflections, etc.
    • Uses constant memory for accelerating access to a list of objects that never changes.
• heat_transfer
  • A simple (physically inaccurate) example of heat transfer in a grid.
    • Uses texture memory for accelerating access to spatially neighboring pixels.

Yes, every code sample here is something simple!

How to run the project

Every code sample folder the following structure:

• cuda folder - contains all the CUDA code;
• lib folder - will contain .lib files with compiled CUDA code;
• build.bat - script containing instructions for building the sample (compilation, linking);
• clear.bat - a helper script to remove all files produced after building;
• setup.py - contains instructions for the Cython compiler on how to make the Python extension with CUDA;
• test.py - contains code for testing the CUDA extension;
• wrapper.pyx - Cython wrapper code around the CUDA code.

To run the sample, complete the following steps:

1. Make sure you have CUDAHOME environment variable which contains path to the CUDA Toolkit folder. Example: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2.
2. Make sure you have installed python packages listed in requirements.txt.
3. Run build.bat.
4. Run test.py.

Tested hardware

All the code was built and tested with the following setup:

• OS: Windows 10.0.19041
• CPU: Ryzen 7 4800H
• GPU: RTX2060 6GB
• CUDA Version: 11.2
• GPU Driver Version: 462.80
• Compiler: Microsoft Visual Studio Community 2019