Python for High Performance Computing
=====================================


Aim
---

Python is an increasingly popular choice for developing scientific
code, mainly owing to its ease of use and availability of libraries
for scientific computing.

This course explores some of the solutions the Python programmer needs
to adopt in order to obtain the code performance required for large
scientific computations.  First, the popular NumPy and SciPy libraries
are revised, followed by an introduction to coupling Python with
compiled language programming (C and Fortran) in order to accelerate
performance critical code.  Then, some of the options available for
parallel programming, in order to exploit the availability of multiple
cores on modern hardware and further improve code performance.


Prerequisites
-------------

This course will assume a level understanding of fundamental
programming concepts (variables, loops, conditionals, functions as
well as fundamentals of object-oriented programming).

It will also assume familiarity with basic Python syntax, along with
data structures such as lists and tuples.

The course will start with a very brief recapitulation of basic Python
features.  For less experienced programmers, however, a concise
introductory online course on python programming is available from
https://www.codecademy.com .


Software
--------

The course is based on Python 2.7.  We strongly the Anaconda
distribution available from

https://www.continuum.io/downloads

This has all the packages that will be required for the main part of
the course: Numpy, SciPy, Matplotlib, jupyter.

In addition to that, the course needs

   * a C and a Fortran compiler (gcc and gfortran are recommended)
   * a MPI library
   * the mpi4py Python module


Contents
--------

The main contents is in the directories

   images     ... contains images used by all notebooks
   notebooks  ... contains the jupyter notebooks for the course
   practicals ... contains the practicals associated with the course

Additional material is in

   additional
   python_hpc_artwork.pptx


Course notebooks
----------------

The notebooks are all in the directory notebooks.  They are

   lecture00-intro             ... course introduction (general)
   lecture01-python            ... introducing python
   lecture02-numpy             ... introducing numpy
   lecture03-matplotlib        ... introducing matplotlib
   lecture04-scipy             ... introducing scipy
   lecture05-extensions        ... (serial) C/Fortran extensions
   lecture06-cython            ... introducing cython
   lecture07-summary-first     ... summary of serial solutions tested on the 2D heat equation
   lecture08-intro-parallel    ... course introduction (to parallel computing)
   lecture09-extensions-openmp ... (parallel) C/Fortran extensions multithreaded using OpenMP
   lecture10-numba             ... introducing numba
   lecture11-summary-second    ... summary of multithreaded solutions tested on the 2D heat equation
   lecture12-multiprocessing   ... introducing multiprocessing
   lecture13-mpi4py            ... introducing mpi4py
   lecture14-conclusion        ... course wrap-up


The Lectures
------------
Lectures are presented as jupyter notebooks.  To run the notebooks, go to the topmost directory and run

   jupyter notebook

or

   jupyter-notebook

To choose a particular browser, the command is

   jupyter notebook --browser=firefox

Use the web bowser to navigate to individual lectures (e.g. "lecture12-multiprocessing") and start the presentations.