GPU-enhanced Neuronal Networks (GeNN)

GeNN is a GPU-enhanced Neuronal Network simulation environment based on code generation for Nvidia CUDA.

Installation

You can download GeNN either as a zip file of a stable release, checkout the development version using the Git version control system or use our Docker container.

Downloading a release

Point your browser to https://github.com/genn-team/genn/releases and download a release from the list by clicking the relevant source code button. After downloading continue to install GeNN as described in the GitHub installing section below.

Obtaining a Git snapshot

If it is not yet installed on your system, download and install Git (http://git-scm.com/). Then clone the GeNN repository from Github

git clone https://github.com/genn-team/genn.git

The github url of GeNN in the command above can be copied from the HTTPS clone URL displayed on the GeNN Github page (https://github.com/genn-team/genn).

This will clone the entire repository, including all open branches. By default git will check out the master branch which contains the source version upon which the next release will be based. There are other branches in the repository that are used for specific development purposes and are opened and closed without warning.

Installing GeNN

Installing GeNN comprises a few simple steps ¹ to create the GeNN development environment:

If you have downloaded a zip file, unpack GeNN.zip in a convenient location. Otherwise enter the directory where you downloaded the Git repository.
Add GeNN's 'bin' directory to your path, e.g. if you are running Linux or Mac OS X and extracted/downloaded GeNN to $HOME/GeNN, this can be done with:
```
export PATH=$PATH:$HOME/GeNN/bin
```
to make this change persistent, this can be added to your login script (e.g. .profile or .bashrc) using your favourite text editor or with:
```
echo "export PATH=$PATH:$CUDA_PATH/bin" >> ~/.bash_profile
```
If you are using Windows, the easiest way to modify the path is by using the 'Environment variables' GUI, which can be accessed by clicking start and searching for (by starting to type) 'Edit environment variables for your account'. In the upper 'User variables' section, scroll down until you see 'Path', select it and click 'Edit'. Now add a new directory to the path by clicking 'New' in the 'Edit environment variable' window e.g.: if GeNN is installed in a sub-directory of your home directory (%USERPROFILE% is an environment variable which points to the current user's home directory) called genn.
Install the C++ compiler on the machine, if not already present. For Windows, download Microsoft Visual Studio Community Edition from https://www.visualstudio.com/en-us/downloads/download-visual-studio-vs.aspx. When installing Visual Studio, one should select the 'Desktop development with C++' configuration. Mac users should download and set up Xcode from https://developer.apple.com/xcode/index.html , Linux users should install the GNU compiler collection gcc and g++ from their Linux distribution repository, or alternatively from https://gcc.gnu.org/index.html
If your machine has a GPU and you haven't installed CUDA already, obtain a fresh installation of the NVIDIA CUDA toolkit from https://developer.nvidia.com/cuda-downloads Again, be sure to pick CUDA and C++ compiler versions which are compatible with each other. The latest C++ compiler need not necessarily be compatible with the latest CUDA toolkit.
GeNN uses the CUDA_PATH environment variable to determine which version of CUDA to build against. On Windows, this is set automatically when installing CUDA. However, if you choose, you can verify which version is selected by looking for the CUDA_PATH environment variable in the lower 'System variables' section of the GUI you used to configure the path: here, CUDA 10.1 and 11.4 are installed and CUDA 11.4 is selected via CUDA_PATH. However, on Linux and Mac you need to set CUDA_PATH manually with:
```
export CUDA_PATH=/usr/local/cuda
```
assuming CUDA is installed in /usr/local/cuda (the standard location on Ubuntu Linux). Again, to make this change persistent, this can be added to your login script (e.g. .profile or .bashrc)

This normally completes the installation. Windows users must close and reopen their command window so changes to the path take effect.

If you are using GeNN in Windows, the Visual Studio development environment must be set up within every instance of the CMD.EXE command window used. One can open an instance of CMD.EXE with the development environment already set up by navigating to Start - All Programs - Microsoft Visual Studio - Visual Studio Tools - x64 Native Tools Command Prompt. You may also wish to create a shortcut for this tool on the desktop, for convenience.

Docker

You can also use GeNN through our CUDA-enabled docker container which comes with GeNN pre-installed. To work with such CUDA-enabled containers, you need to first install CUDA on your host system as described above and then install docker and the NVIDIA Container Toolkit as described in https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker. You can then build the GeNN container yourself or download it from Dockerhub.

Building the container

The following command can be used from the GeNN source directory to build the GeNN container:

make docker-build

This builds a container tagged as genn:latest so, to use this container rather than downloading the prebuild one from dockerhub, just replace gennteam/genn:latest with genn:latest in the following instructions. By default, the container image is based off the Ubuntu 20.04 image with CUDA 11.5 provided by NVIDIA but, if you want to use a different base image, for example to use the container on a machine with an older version of CUDA, you can invoke docker build directly and specify a different tag (listed on https://gitlab.com/nvidia/container-images/cuda/blob/master/doc/supported-tags.md) via the BASE build argument. For example to build using CUDA 11.3 you could run:

docker build  --build-arg BASE=11.3.0-devel-ubuntu20.04 -t genn:latest_cuda_11_3 .

Interactive mode

If you wish to use GeNN or PyGeNN interactively, you can launch a bash shell in the GeNN container using the following command:

docker run -it --gpus=all gennteam/genn:latest

You can also provide a final argument to launch a different executable e.g. /bin/sh to launch a dash shell. NOTE PyGeNN is installed in the system Python 3 environment, the interpreter for which is launched with python3 (rather than just python) on Ubuntu 20.04.

Accessing your files

When using the GeNN container you often want to access files on your host system. This can be easily achieved by using the -v option to mount a local directory into the container. For example:

docker run -it --gpus=all -v $HOME:/local_home gennteam/genn:latest

mounts the local user's home directory into /local_home within the container. However, all of the commands provided by the GeNN container operate using a non-elevated, internal user called 'genn' who, by default, won't have the correct permissions to create files in volumes mounted into the container. This can be resolved by setting the LOCAL_USER_ID and LOCAL_GROUP_ID environment variables when running the container like:

docker run -it --gpus=all -e LOCAL_USER_ID=`id -u $USER` -e LOCAL_GROUP_ID=`id -g $USER` -v $HOME:/local_home gennteam/genn:latest

which will ensure that that 'genn' user has the same UID and GID as the local user, meaning that they will have the same permissions to access the files mounted into /local_home.

Running Jupyter Notebooks

A Jupyter Notebook environment running in the container can be launched using the notebook command. Typically, you would combine this with the -p 8080:8080 option to 'publish' port 8080, allowing the notebook server to be accessed on the host. By default, notebooks are created in the home directory of the 'genn' user inside the container. However, to create notebooks which persist beyond the lifetime of the container, the notebook command needs to be combined with the options discussed previously. For example:

docker run --gpus=all -p 8080:8080 -e LOCAL_USER_ID=`id -u $USER` -e LOCAL_GROUP_ID=`id -g $USER` -v $HOME:/local_home gennteam/genn:latest notebook /local_home

will create notebooks in the current users home directory.

Running PyGeNN scripts

Assuming they have no additional dependencies, PyGeNN scripts can be run directly using the container with the script command. As scripts are likely to be located outside of the container, the script command is often combined with the options discussed previously. For example, to run a script called test.py in your home directory, the script command could be invoked with:

docker run --gpus=all -e LOCAL_USER_ID=`id -u $USER` -e LOCAL_GROUP_ID=`id -g $USER` -v $HOME:/local_home gennteam/genn:latest script /local_home/test.py

Usage

Sample projects

At the moment, the following C++ example projects are provided with GeNN:

Self-organisation with STDP in the locust olfactory system (Nowotny et al. 2005):
- with all-to-all connectivity, using built-in neuron and synapse models (for benchmarks see Yavuz et al. 2016)
- with sparse connectivity for some synapses, using user-defined neuron-and synapse models (for benchmarks see Yavuz et al. 2016)
- using BITMASK connectivity
- using synapses with axonal delays
Pulse-coupled network of Izhikevich neurons (Izhikevich 2003) (for benchmarks see Yavuz et al. 2016)
Genetic algorithm for tracking parameters in a Hodgkin-Huxley model cell
Classifier based on an abstraction of the insect olfactory system (Schmuker et al. 2014)
Cortical microcircuit model (Potjans et al. 2014)
Toy examples:
- Single neuron population of Izhikevich neuron(s) receiving Poisson spike trains as input
- Single neuron population of Izhikevich neuron(s) with no synapses
- Network of Izhikevich neurons with delayed synapses

In order to get a quick start and run one of the the provided example models, navigate to one of the example project directories in the userproject sub-directory, and then follow the instructions in the README file contained within.

Simulating a new model

The sample projects listed above are already quite highly integrated examples. If you wanted to use GeNN to develop a new C++ model, you would do the following:

The neuronal network of interest is defined in a model definition file, e.g. Example1.cc.
Within the the model definition file Example1.cc, the following tasks need to be completed:
1. The GeNN file modelSpec.h needs to be included,
```
#include "modelSpec.h"
```
2. The values for initial variables and parameters for neuron and synapse populations need to be defined, e.g.
```
NeuronModels::PoissonNew::ParamValues poissonParams(
10.0);      // 0 - firing rate
```
  would define the (homogeneous) parameters for a population of Poisson neurons ².
  
  If heterogeneous parameter values are required for a particular population of neurons (or synapses), they need to be defined as "variables" rather than parameters. See the User manual for how to define new neuron (or synapse) types and the Variable initialisation section for more information on initialising these variables to hetererogenous values.
3. The actual network needs to be defined in the form of a function modelDefinition ³, i.e.
```
void modelDefinition(ModelSpec &model); 
```
4. Inside modelDefinition(), The time step DT needs to be defined, e.g.
```
model.setDT(0.1);
```
  \note All provided examples and pre-defined model elements in GeNN work with units of mV, ms, nF and uS. However, the choice of units is entirely left to the user if custom model elements are used.
MBody1.cc shows a typical example of a model definition function. In its core it contains calls to ModelSpec::addNeuronPopulation and ModelSpec::addSynapsePopulation to build up the network. For a full range of options for defining a network, refer to the User manual.
The programmer defines their own "simulation" code similar to the code in MBody1Sim.cc. In this code,
1. They can manually define the connectivity matrices between neuron groups. Refer to the \ref subsect34 section for the required format of connectivity matrices for dense or sparse connectivities.
2. They can define input patterns or individual initial values for neuron and / or synapse variables. \note The initial values or initialisation "snippets" given in the modelDefinition are automatically applied.
3. They use stepTime() to run one time step on either the CPU or GPU depending on the options passed to genn-buildmodel.
4. They use functions like copyStateFromDevice() etc to transfer the results from GPU calculations to the main memory of the host computer for further processing.
5. They analyze the results. In the most simple case this could just be writing the relevant data to output files.

For more details on how to use GeNN, please see documentation.

If you use GeNN in your work, please cite "Yavuz, E., Turner, J. and Nowotny, T. GeNN: a code generation framework for accelerated brain simulations. Scientific Reports, 6. (2016)"

While GeNN models are normally simulated using CUDA on NVIDIA GPUs, if you want to use GeNN on a machine without an NVIDIA GPU, you can skip steps v and vi and use GeNN in "CPU_ONLY" mode. ↩
The number of required parameters and their meaning is defined by the neuron or synapse type. Refer to the User manual for details. We recommend, however, to use comments like in the above example to achieve maximal clarity of each parameter's meaning. ↩
The name modelDefinition and its parameter of type ModelSpec& are fixed and cannot be changed if GeNN is to recognize it as a model definition. ↩

genema/genn