Scribd
Upload
5 views1 page

Bourne Abstract

Gyrominetic simulation python

Uploaded by

sicih95676
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
5 views1 page

Bourne Abstract

Gyrominetic simulation python

Uploaded by

sicih95676
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Parallel gyrokinetic simulations with Python

Emily Bourne1 , Yaman Güçlü2

1
Technische Universität München, Boltzmannstr. 3, 85748 Garching, Germany
2
Max-Planck-Institut für Plasmaphysik, Boltzmannstr. 2, 85748 Garching, Germany

In the field of scientific computing, two competing requirements constitute a challenge


for efficient code development: on the one hand, algorithm prototyping needs to be fast,
flexible and interactive; on the other hand, the target application often requires single-
core optimization, shared-memory and MPI parallelization, and strict quality control.
As a result, the so-called ‘prototyping’ and ‘production’ phases tend to be completely
separate, using different programming languages and libraries, input/output facilities,
unit tests and diagnostics. Moreover, once the production phase is initiated, it is very
hard to make changes to the code architecture or the basic algorithms. For all these
reasons, the transition from prototyping to production is a large burden, especially for
small teams with limited resources.

In an attempt to smoothen such a transition, we have investigated the possibility of using


Python for the whole development process, and of accelerating the most computationally
intensive kernels with Fortran. The novelty of this approach is that the Python-to-Fortran
translation is performed automatically by Pyccel, a Python 3 acceleration library [1] to
which we provide static type information in the form of pragmas or function decorators.
This allows us to maintain only one version of the code, which can be run as pure
Python or in ‘Fortran-accelerated’ mode. The correctness of the Fortran modules is
simply verified by the original suite of unit tests, and because the Fortran source code is
available, further manual optimization is possible.

For a case study we have taken the semi-Lagrangian solution of the reduced (µ = 0)
gyrokinetic Vlasov equation in screw-pinch geometry, for which a pure Fortran simula-
tion already exists within the Selalib library [2, 3], and we have developed PyGyro, a
Python code that offers small but significant algorithmic improvements over the original
simulation. After acceleration with Pyccel, an ion temperature gradient test-case with
PyGyro runs approximately 20% slower than the original code on 1 node (32 processes),
but can be distributed over a much larger number of nodes (128 instead of just 8). In
this talk we will present PyGyro and its new parallelization algorithms, we will critically
review the obtained performance, and we will discuss future improvements.

References

[1] PYCCEL. https://github.com/pyccel/pyccel


[2] SELALIB. http://selalib.gforge.inria.fr
[3] G. Latu, M. Mehrenberger, Y. Güçlü, M. Ottaviani, E. Sonnendrücker,
J. Sci. Comput. 74 (2018) 1601-1650

You might also like