pyus 0.2.2

Requirements

PyUS is based on Python>=3.5, with the following dependencies:

• numpy

• scipy

• matplotlib

• tqdm

• pycuda

• h5py

• pyfftw

These packages will be installed when calling the pip installation command.

Important notes:

• CUDA 10.0 is required.

• FFTW3 is required

• CPU-only implementations are also provided, though CUDA 10.0 should still be installed in order to work.

• Only tested on Ubuntu 16.04 and 18.04 based distributions, although it should work on any GNU/Linux distribution that uses glibc >= 2.12.

Installation

PyUS is currently under development, and beta versions will soon be available for installation from the Python Package Index (PyPI).

For now, the easiest way to install it is by downloading the pre-compiled wheel contained in this archive, and install it using pip in a dedicated Python environment:

pip install --upgrade pip
pip install pyus

Build from sources

Before building the wheel you will need to build the pulse library. To do so, refer to the pulse_directory.

In order to build the wheel from sources, you will first have to install numpy and cython.

pip install numpy cython

An option to provide for the build is the CUDA version that should be used. To do so, just pass it as an environment variable during the call to pip:

CUDA_VERSION=X.Y pip <pip arguments>

If you don’t provide it, the version will be automatically detected on your system, provided that CUDA is installed in the standard location (i.e. /usr/local/). If you have several versions installed or none, an error will be raised and the build will be interrupted. You will then have to provide a specific version of CUDA or install one.

To build the package, you have several options:

• you can manually build the wheel:

  cd pyus
  CUDA_VERSION=X.Y pip wheel --wheel-dir=<your/wheel/dst/dir> --no-deps .

  where the --wheel-dir option allows you to specify the directory to put the built wheel in, and the --no-deps prevents pip from downloading all the dependencies and putting them with your built wheel.

• you can install the package locally in editable mode:

  cd pyus
  CUDA_VERSION=X.Y pip install -e .

• you can install the package normally:

  cd pyus
  CUDA_VERSION=X.Y pip install .

Getting Started

As PyUS is initially an internal package used for research purposes, it still lacks a lot of documentation. We did our best to use type hinting that “speak for themselves”.

In the meantime, some examples are provided under directory examples/.

Contributing

See CONTRIBUTING.rst.

Team/Contributors

TODO

License

PyUS is released under the terms of the BSD 3-Clause license.

Acknowledgments

PyUS is initially an internal research package that was started in 2018 at the Signal Processing Laboratory 5 (LTS5), École polytechnique fédérale de Lausanne.

TODO: citation