Section: New Results

BOAST: a Metaprogramming framework for computing kernels

Participants : Brice Videau [Postdoc CNRS, Mont-Blanc] , Kevin Pouget [UJF, Nano2017] , Luigi Genovese [Researcher, CEA INAC] , Thierry Deutsch [Researcher, CEA INAC] , Anthony Leonard [CNRS, Polytech Grenoble, Internship, from May 2015 until Aug 2015] , Frederic Desprez, Jean François Méhaut.

Porting and tuning HPC applications to new platforms is tedious and costly in terms of human resources. Nonetheless, it is a very important aspect of the Mont-Blanc project. Indeed, for the Mont-Blanc project, more than ten applications were selected to be ported and optimized for the prototype platform.

Unfortunately, portability efforts are often lost when migrating to a new architecture. Worse, code may lose maintainability because several versions of some functionalities coexist, usually with a lot of duplication. Thus productivity of porting and tuning efforts is low as a huge fraction of those developments are never used after the platform they were intended for is decommissioned. Genericity of HPC codes is often limited. One of the reason is that producing generic code in Fortran 90/95 is difficult as the language does not really fit for it. Sometimes, adding genericity degrades performance as optimization opportunities that come from over-specification are lost. Functionality of HPC codes is tied to the previous point. Without genericity, adding new functionalities can be quite costly.

BOAST is a metaprogramming framework to produce portable and efficient computing kernels for HPC application. BOAST offers an embedded domain specific language to describe the kernels and their possible optimization. BOAST also supplies a complete runtime to compile, run, benchmark, and check the validity of the generated kernels. BOAST is being used in two flagship HPC applications BigDFT and SPECFEM3D, to improve performance portability of those codes.

BOAST is developped in the context of Mont-Blanc projects. It will be also used in the C2S@Exa IPL and the H2020/HPC4E project.