Porting TELEMAC-MASCARET to OPENPOWER and experimenting GPU offloading to accelerate the TOMAWAC module

In this paper the state of porting TELEMACMASCARET
on the OPENPOWER architecture with different
compilers is shown. A port to GPUs with OpenMP and
OpenACC of a computationally intensive subroutine of
TOMAWAC is also explained and the performance benefits
shown, a comparison with an x86-64 machine is also presented.
Finally ongoing work is presented and discussed: the port of a
complete and more challenging test-case, a triple coupling case
from EDF.