Pavanakumar Mohanamuraly

Senior Researcher

CERFACS, Toulouse, France

Biography

I am a senior researcher at ALGO-COOP team, CERFACS, Toulouse, France. My interests are in the area of uncertainty quantification, optimisation, and computational fluid dynamics. At CERFACS I develop and maintain the TreePart/Adapt framework for large scale online unstructured mesh partitioning and adaptation. We recently achieved scalable online partitioning up to 132k MPI ranks fully exploiting the hardware hierarchy of the parallel machines (node/sockets/cores).

Am also the developer and maintainer of the Turbomachinery CFD/Adjoint solver (STAMPS) along with Dr. Jens-Dominik Mueller at Queen Mary University of London. STAMPS is one of the first Turbomachinery RANS solvers based on Newton-Krylov method using exact Jacobian obtained using source-transformed Algorithmic differentation.

Feel free to explore my open source projects and the most starred one on GitHub the compressibleFoam solver (an excellent CFD teaching aid) and my blog posts.

Interests

Uncertanity quantification
Adjoint methods
Aero-acoustic design

Education

PhD in Aerospace Engineering, 2019

Queen Mary University of London
MS in Aerospace Engineering, 2010

Pennsylvania State University
B.Tech in Aerospace Engineering, 2005

Madras Institute of Technology

Publications

More Publications

M Pavanakumar, J D Mueller. A meshless optimised mesh-smoothing framework. In NOED conference, 2016.

PDF Code Dataset Project Slides Video Custom Link

M Pavanakumar, Jan Huckelhiem, Jen Mueller. Hybrid Parallelisation of an Algorithmically Differentiated Adjoint Solver. In ECCOMACS conference, 2016.

PDF Code Dataset Project Slides Video Custom Link

Recent Posts

Wrapping Fortran Derived Types : Part 2

May 24, 2019 1 min read

In a previous post wrapping user-defined types in Fortran into C was explained using the opaque object hack. The hack was improved by adding portable size detection and direct access to array memory using pointer handles. For compilers that do not yet support STORAGE_SIZE intrinsic function an alternative route is suggested in this post. Again the approach is based on the one by Pletzer et. al., “Exposing Fortran derived types to C and other languages”, 2008 but modified and improved to increase portability and safety.

Wrapping Fortran Derived Types : Part 1

May 23, 2019 7 min read

Abstract Fortran C/C++/Python interoperability is an important consideration while designing scientific HPC codes. Fortran to a large extent still dominates in terms of lines of code in scientific software due to its legacy, reliable mathematical libraries and many years of research into the design of its compilers. Therefore it becomes mandatory to support an array of codes and libraries developed in mixed languages say Fortran, C and C++. It is counter productive to rewrite the codes and revalidate just for the sake of using a single monolithic language.