Brian2CUDA is an extension of the spiking neural network simulator Brian2, implementing a Brian2 standalone device to generate C++/CUDA code to run simulations on NVIDIA general purpose graphics processing units (GPGPUs).

Use your Brian2 code (see Brian2 documentation) and modify the imports to:

from brian2 import *
import brian2cuda
set_device("cuda_standalone")

Only Linux environments are supported. For progress on Windows support, see #225.

The correct Brian2 version with which this implementation is working is stored in a submodule in frozen_repos/brian2

After cloning Brian2CUDA, you can initialize the submodule from inside this repository:

cd brian2cuda
git submodule update --init frozen_repos/brian2

Next, you need to incorporate a few changes into the Brian2 repository. These changes are already included in the newest Brian2 version, but need to be added manually to the Brian2 version that is used in frozen_repos. To add these changes, apply the brian2.diff file to the repository:

cd frozen_repos
bash update_brian2.sh
# return to the brian2cuda root directory for the following installation
cd ..

Now you can install the correct Brian2 and Brian2CUDA versions using pip (Be careful if you already have a Brian2 version installed in your current Python environment!):

pip install .
pip install ./frozen_repos/brian2

If you want to install in developer mode, add the -e flag to the pip commands.

Or just add them to your PYTHONPATH (in which case you need to install their dependencies manually).

The Brian2GeNN and GeNN versions that can be used with the same Brian2 version as Brian2CUDA are stored in submodules in frozen_repos/brian2genn and frozen_repos/genn. You can initialize them with:

git submodule update --init frozen_repos/brian2genn
git submodule update --init frozen_repos/genn

Now you can install Brian2GeNN either with pip:

pip install ./frozen_repos/brian2genn

or just add it to your PYTHONPATH.

For GeNN to work correctly, you need to set the environment variable GENN_PATH. You can use the following command:

source frozen_repos/init_genn.sh

If you are using a conda environment, make sure that you don't have any compiler packages installed in your environment or unset the corresponding environment variables. Otherwise you can get mismatches of compiler and linker versions in GeNN. During the first GeNN run, a set of libraries are compiled once. If you change you compiler or environment later on, you might have to reinstall GeNN (remove the compiled filed).

Otherwise follow the instructions from the GeNN repository.