.. image:: https://travis-ci.org/NOhs/pytest-easyMPI.svg?branch=master :target: https://travis-ci.org/NOhs/pytest-easyMPI

.. image:: https://badge.fury.io/py/pytest-easyMPI.svg :target: https://badge.fury.io/py/pytest-easyMPI

.. image:: https://img.shields.io/badge/License-MIT-blue.svg :target: https://opensource.org/licenses/MIT

.. image:: https://app.codacy.com/project/badge/Grade/23f4495e7d19402f93aa29b92885f281 :target: https://www.codacy.com/gh/NOhs/pytest-easyMPI/dashboard?utm_source=github.com&utm_medium=referral&utm_content=NOhs/pytest-easyMPI&utm_campaign=Badge_Grade

Welcome to the pytest-easyMPI package!

This package aims at making MPI code testing as similar to testing regular serial code as possible. In doing so hopefully the users of this plugin can focus more on writing tests of their MPI code and spend less time figuring out how to integrate MPI tests into their other test cases.

Short example

The following shows an example of how to combine a serial and a parallel test in a single test file. The parallel test is run using 4 MPI ranks:

.. code:: python

from pytest_easyMPI import mpi_parallel

def test_serial():
    assert True

@mpi_parallel(4)
def test_parallel():
    # Import MPI only inside the test that needs it
    # (to avoid spawning too many MPI communicators)
    from mpi4py import MPI

    data = MPI.COMM_WORLD.gather(MPI.COMM_WORLD.Get_rank())
    if MPI.COMM_WORLD.Get_rank() == 0:
        assert sum(range(MPI.COMM_WORLD.Get_size())) == sum(data)

The test can then be run by calling::

pytest