FLEXIBLE DATA STRUCTURES FOR SCALABLE CFD CODES ON EMERGING ARCHITECTURES

Overview & Synopsis

Supercomputing hardware is becoming increasingly diversified. Multi-physics Computational Fluid Dynamics (CFD) codes must adapt to these new heterogeneous machines if large simulations are to be performed efficiently. We propose a flexible and lightweight graph-like data structure capable of decoupling different needs of the code, such as cache blocking and linear system solution. Furthermore this data structure can also improve parallel performances reducing communication time through multi-layer ghosts.