/cs149-local-dev

Provides `conda` installation instructions for Stanford's CS149 (Parallel Computing) programming assignments

MIT LicenseMIT

cs149-local-dev

Recently Stanford published all video recordings from the Fall 2023 CS149 (Parallel Computing) course. It is an exellent resource for learning about parallel computing on CPUs and GPUs, and hardware specific programming languages like CUDA and ISPC.

It appears that the course offers allocation of AWS instances and access to Stanford's Myth machines. This repository is an attempt to provide instructions for setting up a local development environment for the course programming assignments.

Some important notes:

  • The assignments occasionally provide local installation instructions, but they are not always cross-platform. Always check first the original assignment repositories for the most up-to-date instructions. This repository is not affiliated with Stanford University.
  • About half of Assignment 1 involves ISPC code. Check the ISPC website to verify that your system is compatible. It appears that while ISPC was originally developed for Intel CPUs, ISPC is now compatible with AMD and ARM-based CPUs as well, including Apple's ARM-based M chips.
  • Assignments 3 and 4 involve CUDA code. You will need an NVIDIA GPU to complete these assignments.
  • These instructions have been tested only on a Ubuntu 22.04 machine with an RTX 3060Ti. If you have tested the instructions on a different system, please submit a pull request to update this README.
  • ✅ - Tested and working
  • 🟡 - Tested and partially working
  • ❔ - Not tested
  • ❌ - Could not get it to work
Assignment Ubuntu 22.04 MacOS 12.3 Windows 10
Assignment 1 ✅ ❔ ❔
Assignment 2 ✅ ❔ ❔
Assignment 3 ✅ ❔ ❔
Assignment 4 ✅ ❔ ❔
Assignment 5 ❔ ❔ ❔

This repository provides conda installation instructions for Stanford's CS149 (Parallel Computing) programming assignments. With conda you can isolate the particular Python environment needed for each assignment (or the course as a whole), and avoid conflicts with other projects on your system.

Getting Started

  • Verify that gcc and g++ are both installed on your system and that their versions match.

    gcc --version
g++ --version

    If they do not match, follow this StackOverflow post to create the appropriate symlinks. Tested with gcc/g++ version 12.3.0 on Ubuntu 22.04.

  • Create a new conda environment for the course:

    conda create -n cs149
conda activate cs149

Assignment 1

Link to Official Github Repo: Assignment 1

Clone the repository and install the dependencies:

git clone git@github.com:stanford-cs149/asst1.git
cd asst1

ISPC: On Ubuntu the prefered method of installation is through the Snap store. Alternatively, with the newly created cs149 conda environment activated, install ispc from the conda-forge channel:

conda install conda-forge::ispc

That's it! You should now be able to build the executables for each part of the assignment by running make in the respective directories.

Note

In Program 6 (K-Means) you need to uncomment the portion of main.cpp that generates the input data. After you have generated the input data, you can comment out that portion of the code and run the program as normal.

Assignment 2

Link to Official Github Repo: Assignment 2

No additional dependencies are required for this assignment (a compatible g++ is all we need).

Clone the repository and build the executables for each part of the assignment by running make in the respective directories. The included executable part_a/runtasks_ref_linux also was tested and it works.

Assignment 3

Link to Official Github Repo: Assignment 3

Clone the repository via git as in the previous assignments.

Note

This assignment requires a compatible NVIDIA GPU and CUDA.

Note

As mentioned in one of the provided readme here, if you are on an Ubuntu 22.04 machine, you may be able to install everything via the install.sh script provided in the assignment repository:

chmod +x ./install.sh
./install.sh

However that this script is intended for the AWS VMs, so use it at your own risk.

With conda, we can install CUDA via NVIDIA's official conda channel. However, this installation defines environment variables such as CXX and CXXFLAGS that are also referenced in the Makefile of the assignment. Getting the Makefile to work with the conda-installed CUDA and OpenGL+Glut appears to be very tricky.

A solution is to install cuda system-wide via the installer provided by NVIDIA for CUDA Toolkit. For the rendering part, GLUT can be installed in Ubuntu via apt:

sudo apt-get install freeglut3-dev

Finally you might need to specify the compute capability of your GPU in the Makefile of the assignment. Specifically for GeForce 30 series GPUs, you need to specify -arch=sm_86 in the NVCCFLAGS variable in the Makefile:

NVCCFLAGS = -O3 -arch=sm_86

You can find the compute capability of your GPU here.

Assignment 4

After a bit of trial and error, the following combination of package versions appears to work on Ubuntu 22.04:

conda create -n gpt149
conda activate gpt149
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=11.8 python=3.10 numpy=1.26 ninja tiktoken -c pytorch -c nvidia -c conda-forge