Section: New Software and Platforms

BOAST

Bringing Optimization Through Automatic Source-to-Source Transformations

Keywords: Code generation - Portability - Autotuning - High performance computing - Conformance testing - Productivity

Functional Description: BOAST provides scientific application developers with a framework to develop and test application computing kernels.

The developer starts from an application kernel (either designed or implemented), and writes it in a dedicated language. This language provides enough flexibility for the kernel to be metaprogrammed with several orthogonal optimizations. From this set of optimizations, possible languages targets, and compilation options, the user can design an optimization space to explore. This optimization space can contain rules to remove infeasible candidates. BOAST provides the mechanisms to specify those optimization spaces and enforce the users rules.

BOAST was already used with three real scientific applications: BigDFT (materials, CEA Inac), SPECFEM3D (geophysics, CNRS and Princeton) and GYSELA (plasma physics, CEA Cadarache, ITER). '

• Partner: CEA INAC LSim

• Contact: Brice Videau

• URL: https://github.com/Nanosim-LIG/boast