Python Packaging

PYNQ uses pip - the Python Packaging Authority’s recommended Python Package Installer to install and deliver custom installations of PYNQ. pip’s flexible package delivery model has many useful features.

Packaging pynq for Distribution

Packaging the pynq Python project that pip can use is hugely beneficial, but requires careful thought and project architecture. There are many useful references that provide up-to-date information. For more information about how the pynq library is packaged see the following links:

Delivering Non-Python Files

One extremely useful feature that pip provides is the ability to deliver non-python files. In the PYNQ project this is useful for delivering FPGA binaries (.bit), overlay metadata files (.hwh), PYNQ MicroBlaze binaries (.bin), and Jupyter Notebooks (.ipynb), along side the pynq Python libraries. The most straightforward way of including non-python files is to add a to the project.

In addition PYNQ provides two mechanisms that can be used to aid deployments of notebooks and large bitstreams - in particular xclbin files which can exceed 100 MBs each.

Registering PYNQ Notebooks

If you have notebooks in your package, you can register your notebooks with the pynq get-notebooks command line tool by creating a pynq.notebooks entry point linking to the part of your package. The key part of the entry point determines the name of the folder that will be created in the notebooks folder and all of the files in the corresponding package will be copied into it. Any .link files described below will also be resolved for the currently active device.

Example Setup Script

An example of using pip’s file which delivers xclbin files and notebooks using the PYNQ mechanisms is show below.

from setuptools import setup, find_packages
from pynq.utils import build_py
import new_overlay

   name = "new_overlay",
   version = new_overlay.__version__,
   url = '',
   license = 'All rights reserved.',
   author = "Your Name",
   author_email = "",
   packages = find_packages(),
       'pynq.notebooks': [
           'new-overlay = new_overlay.notebooks'
   cmdclass={'build_py': build_py},
   description = "New custom overlay"

A corresponding to add the notebooks and bitstreams files would look like

recursive-include new_overlay/notebooks *
recursive-include new_overlay *.bit *.hwh

If you want to have users be able to install your package without first installing PYNQ, you will also need to create a pyproject.toml file as specified in PEP 518. This is used to specify that PYNQ needs to be installed prior to the setup script running so that pynq.utils.build_py is available for importing. The setuptools and wheel are required by the build system so we’ll add those to the list as well.

requires = ["setuptools", "wheel", "pynq>=2.5.1"]

Registering Overlays Path

If you have overlays in your package, and you want to reference them without the absolute path. You can register the absolute path in in the by creating a pynq.overlays entry point. The key part of the entry point specifies the name of the folder where the overlays will be located.

An example is provided below:

    'pynq.overlays': ['new-overlay = new_overlay.overlays']

Rebuilding PYNQ

Starting from image v2.5, the official PYNQ GitHub repository will not version-control the following files anymore:

  • overlay files (e.g., base.bit, base.hwh),
  • bsp folders(e.g., bsp_iop_pmod)
  • MicroBlaze binaries (e.g., pmod_adc.bin)

We refrain from keeping track of these large files; instead, we rely on the SD build flow to update them automatically in each build. Some side-effects are shown below:

  • Users should expect longer SD image building time when users are building the image for the first time. Subsequent builds are much faster.
  • Users will no longer be able to pip install directly from the official PYNQ GitHub repository.

To get those files manually, users can simply run the located at the root of the PYNQ repository (make sure you have the correct version of Xilinx tools beforehand).

Once you have all the files, including the files mentioned above, you can package the entire GitHub repository into a source distribution package. To do that, run

python3 sdist

After this, you will find a tarball in the folder dist; for example, pynq-<release.version>.tar.gz. This is a source distribution so you can bring it to other boards and install it. From a terminal on a board, installing the pynq Python library is as simple as running:

export BOARD=<Board>
export PYNQ_JUPYTER_NOTEBOOKS=<Jupyter-Notebook-Location>
pip3 install pynq-<release.version>.tar.gz

After pip finishes installation, the board must be rebooted. If you are on a board with a PYNQ image (OS: pynqlinux), you are done at this point. If you are not on a PYNQ image (other OS), the above pip3 install is only for the pynq Python library installation; you also need 2 additional services to be started for pynq to be fully-functional.

  • Jupyter notebook service. (Check <PYNQ-repo>/sdbuild/packages/jupyter/ as an example).
  • Clear PL Statefile service. (Check <PYNQ-repo>/sdbuild/packages/clear_pl_statefile for more information).

Using pynq as a Dependency

One of the most useful features of pip is the ability to depend on a project, instead of forking or modifying it.

When designing overlays, the best practice for re-using pynq code is to create a Python project as described above and add pynq as a dependency. A good example of this is the BNN-PYNQ project.

The BNN-PYNQ project is an Overlay that depends on pynq but does not modify it. The developers list pynq as a dependency in the pip configuration files, which installs pynq (if it isn’t already). After installation, the BNN-PYNQ files are added to the installation: notebooks, overlays, and drivers are installed alongside pynq without modifying or breaking the previous source code.

Needless to say, we highly recommend depending on pynq instead of forking and modifying pynq. An example of depending on pynq is shown in the code segment from the previous section.