PyNN on SpiNNaker Extension Development Guide

Follow me on GitHub

This guide is intended to help you set up your development environment so that you can write new extensions to the implementation of SpiNNaker on PyNN. This might include new neural models, or new connectors. Note that this guide is not intended to help you make changes to the implementation itself, although it is conceivable that once your extension is stable, it might find its way in to the core code base.

sPyNNaker C Development Environment

If you are writing new neural models, synapse models or plasticity models, it is likely that you will need to compile C code to run on SpiNNaker. Firstly, you must follow the instructions to install a C development environment. You should ensure that you install both the spinn_common library and the SpiNNFrontEndCommon optional dependencies. If your code is then to make use of the sPyNNaker neural modelling code base (which avoids the need to rewrite much of the necessary code):

  1. Download the sPyNNaker source code as a zip or tar.gz
  2. Extract the code to the location of your choice.
  3. Create a new environment variable NEURAL_MODELLING_DIRS which is set to the path of the neural_modelling subfolder of the extracted archive.

sPyNNaker Python Development Environment

In addition to the C code, you will also likely need to write Python code which enables the use of the new models from within PyNN. Other extensions to sPyNNaker might also only require python code changes, such as the development of new connector types. The only requirements for the Python development, is a working sPyNNaker install.

If the code that you are writing only extends the Python functionality, this can be written as with any normal Python code, using sPyNNaker as a library as required. The new code can then be used from PyNN scripts by importing the code separately from sPyNNaker.

sPyNNaker C and Python Project

When new C code is to be written, new Python code will also be required to interface the C code with PyNN scripts. We currently recommend keeping these elements in a single project, as this will help to keep them synchronized.

The recommended structure of a new project is as follows (replacing <module-name> with an appropriate name for your module):

sPyNNaker<module-name>Extension/
    neural_modelling/
        <module-name>/
            Makefile
            ...
    <module-name>/
        model_binaries/
            __init__.py
        __init__.py
        ...
    setup.py