Python for HPC#

Welcome to the MPCDF Python for HPC course!

Authors#

Max Planck Computing and Data Facility, Garching

Course material#

  • This material is currently intended to be used only by the registered participants of a Python for HPC course given by the MPCDF

  • Redistribution requires the consent of the authors

  • A git repository of the material is provided at https://gitlab.mpcdf.mpg.de/mpcdf/python-for-hpc-exercises

  • The Jupyter notebooks discussed during the lectures are bundled into a Jupyter book for convenient reading.

List of topics#

  • Introduction

    • Python refresher

    • Basic HPC

  • Efficient numerical computing

    • NumPy

    • SciPy

    • HDF5-based IO with h5py

    • Cython

    • Interfacing with C/C++, Fortran, CUDA, and libraries

    • JIT, Numba, Jax

    • Profiling

  • Parallel computing

    • Parallel computing basics

    • Python threading, GIL

    • multiprocessing

    • GPU computing using Numba, CuPy, Jax

    • Parallelization frameworks, e.g. Dask

    • mpi4py

    • Running parallel Python programs with Slurm

  • Software engineering with Python

    • Testing

    • Packaging

    • Documentation

  • Visualization

    • matplotlib

    • Colors

  • Exercises and examples

    • Basic Python

    • NumPy

    • simple advection code

    • simple diffusion code (MPI)

References#

This course is largely based on our experience from daily work. In addition, the following sources were used:

  • High Performance Python, Practical Performant Programming for Humans, Micha Gorelick, Ian Ozsvald, O’Reilly Media; Second Edition, 2020. (In particular, parts of the diffusion example are discussed similarly to the presentation in this book.)

  • A Whirlwind Tour of Python, Jake VanderPlas, O’Reilly Media, 2016.

  • official documentation of Python, NumPy, SciPy, Cython, Numba, mpi4py, etc.

Other (minor) sources are referenced directly in the notebooks.

Software prerequisites#

Python packages#

The examples discussed in this course are based on Python 3 and NumPy, SciPy, Cython, Numba, matplotlib, mpi4py, Dask, and few more.

To conveniently get access to all the required packages, users can download and install Miniforge – which is a free alternative to commercial Python distributions – and use conda or mamba together with the file environment.yml from this repository to create a local software environment.

Jupyter presentation via RISE#

  • Presentation of the Jupyter notebook cells as slides is possible via the RISE extension.

    • To enter presentation mode, simply press <Alt+r> from within a notebook.