Arbor’s Python API will be the most convenient interface for most users.
Arbor requires Python version 3.6 and later. It is advised that you update pip as well.
Every point release of Arbor is pushed to the Python Package Index. The easiest way to get Arbor is with pip:
pip3 install arbor
You will need to have some development packages installed in order to build Arbor this way.
Ubuntu/Debian: sudo apt install git build-essential python3-dev python3-pip libxml2-dev
Fedora/CentOS/Red Hat: sudo yum install git @development-tools python3-devel python3-pip libxml2-devel
macOS: get brew here and run brew install cmake clang python3 libxml2
Windows: the simplest way is to use WSL and then follow the instructions for Ubuntu.
If you wish to get the latest Arbor straight from the master branch in our git repository, you can run:
pip3 install git+https://github.com/arbor-sim/arbor.git
To test that Arbor is available, try the following in a Python interpreter to see information about the version and enabled features:
>>> import arbor >>> print(arbor.__version__) >>> print(arbor.__config__)
To get help in case of problems installing with pip, run pip with the
--verbose flag, and attach the output
(along with the pip command itself) to a ticket on Arbor’s issues page.
If you want to work on Arbor’s code, you can get a copy of our repo and point pip at the local directory:
# get your copy of the Arbor source git clone https://github.com/arbor-sim/arbor.git --recursive # make your changes and then instruct pip to build and install the local source pip3 install ./arbor/
Every time you make changes to the code, you’ll have to repeat the second step.
By default Arbor is installed with multi-threading enabled. To enable more advanced forms of parallelism, the following optional flags can be used to configure the installation:
--mpi: Enable MPI support (requires MPI library).
--gpu: Enable GPU support for NVIDIA GPUs with nvcc using
cuda, or with clang using
cuda-clang(both require cudaruntime). Enable GPU support for AMD GPUs with hipcc using
hip. By default set to
none, which disables gpu support.
--vec: Enable vectorization. This might require choosing an appropriate architecture using
--arch. Note that on x86-64 platforms compilation will fail if you enable vectorization, but the CPU or
--archdoes not support any form of AVX.
--arch: CPU micro-architecture to target. By default this is set to
setup.py the flags must come after
install on the command line,
and if being passed to pip they must be passed via
--install-option. The examples
below demonstrate this for both pip and
Vanilla install with no additional features enabled:
pip3 install arbor python3 ./arbor/setup.py install
With MPI support. This might require loading an MPI module or setting the
pip3 install --install-option='--mpi' ./arbor python3 ./arbor/setup.py install --mpi
pip3 install --install-option='--vec' --install-option='--arch=skylake' arbor python3 ./arbor/setup.py install --vec --arch=skylake
Enable NVIDIA GPUs (compiled with nvcc). This requires the CUDA toolkit:
pip3 install --install-option='--gpu=cuda' ./arbor python3 ./arbor/setup.py install --gpu=cuda
Enable NVIDIA GPUs (compiled with clang). This also requires the CUDA toolkit:
pip3 install --install-option='--gpu=cuda-clang' ./arbor python3 ./arbor/setup.py install --gpu=cuda-clang
Enable AMD GPUs (compiled with hipcc). This requires setting the
pip3 install --install-option='--gpu=hip' ./arbor python3 ./arbor/setup.py install --gpu=hip
Setuptools compiles the Arbor C++ library and
wrapper, which can take a few minutes. Pass the
--verbose flag to pip
to see the individual steps being performed if you are concerned that progress
If a downstream dependency requires Arbor be built with
a specific feature enabled, use
define the constraints.
For example, a package that depends on arbor version 0.3 or later
with MPI support would add the following to its requirements:
arbor >= 0.3 --install-option='--gpu=cuda' \ --install-option='--mpi'
Note on performance¶
The Python interface can incur significant memory and runtime overheads relative to C++ during the model building phase, however simulation performance is the same for both interfaces.