Fortran to OpenCL parallelization

Master or engineer internship

Keywords: compilation, parallelization, Fortran, OpenCL, program

transformations

Manycore processors and heterogeneous accelerators allow good performance on parallel programs with good power efficiency but their programming is quite challenging.

There exist some frameworks to ease application development on these platforms and Par4All (http://par4all.org) is one of them, based on automatic parallelization that can parallelize C and Fortran sequential programs to OpenMP, CUDA and OpenCL.

Unfortunately, Fortran support is quite limited and we focus with this internship on adding more support for Fortran 95 though the GNU/gfortran front-end to PIPS (http://pips4u.org) and restructuring the Par4All core to be more extensible and to deal more languages.

This internship can be followed by a job or a PhD thesis.

Some knowledge useful for this project: compilation, C, Fortran, Python, Unix.

Advisor : Thierry.porcher at hpc-project dot com, http://par4all.org

HPC Project http://www.hpc-project.com is a start-up with around 35 people.

Meudon (92) or Montpellier (34), France.