Sites Inria

Version française



Inria joins OpenMP ARB to take up one of the major challenges of the Exascale era: parallel language standardisation

The Klang-OMP compiler automatically translates an OpenMP compliant application into an application exploiting the targeted runtime system.

On 22 July 2016 Inria joined the OpenMP Architecture Review Board (OpenMP ARB), an association of 28 manufacturers, software publishers and research bodies, all leaders in their field, to collaborate in the design of a standard parallel programming model using shared memory. 

This event occurred while President Obama recently announced the decision to provide unprecedented support to research and development in supercomputing in the US (NSCI). In return, academic and industrial stakeholders located in the US are strongly enjoined to unite their strengths and to base their research around standardised technologies such as OpenMP.

The STORM team (Inria Bordeaux – Sud-Ouest), specialised in the development of programming tools for scientific applications used in high performance computing (HPC) on parallel machines is behind the initiative to join the OpenMP ARB consortium. In fact, "Inria project-teams have been developing runtime systems and compilation techniques for parallel programming for several decades", comments Olivier Aumage, an Inria researcher on the  STORM team, "On joining OpenMP ARB today, Inria  looks forward to contributing this expertise so that OpenMP can respond to the challenges of the Exascale era".

The birth in 2015 of the Klang-OMP compiler adapted to the specifications of the OpenMP parallel programming language — after two years of work by Action de Développement Technologique (ADT), Inria K'Star coordinated by Olivier Aumage (STORM team, Inria Bordeaux - Sud-Ouest) and Thierry Gautier (Avalon team, Inria Grenoble - Rhône-Alpes) — was the defining element in Inria joining OpenMP ARB. This source-to-source compiler, in effect, enables applications written according to the OpenMP standard to benefit immediately and fully from the performance and capacity of runtime systems such as StarPU (developed by the STORM team) and Kaapi (developed by the AVALON team), without any specific adaptation. There is currently no equivalent on the market in terms of portability across different runtime systems, flexibility of execution and demonstrating potential extensions to OpenMP language.

Our participation in the consortium is, initially, an opportunity to offer greater visibility to runtimes developed by Inria, in particular, outside the European context. This collaboration will also facilitate the development of projects as “Project H2020 INTERTWinE” which aims to develop programming model interoperability and raise awareness among groups such as the MPI Forum and OpenMP ARB about the production of modular and inter-compliant standards. Additionally, thanks to the integration of developments due to new results, a large number of applications that use OpenMP could benefit directly from the results of research work by Inria teams. Finally, as Michael Klemm, CEO of OpenMP ARB puts it: "We warmly welcome the arrival of Inria within the OpenMP family. Their long history of innovation and invention will benefit OpenMP."