Enhancing finite element computations on emerging parallel architectures

The project aims to address some of the challenges of using modern parallel hardware, in the context of advanced finite element computaitons. First, we will investigate the suitability of a number of programming models for exploiting the shared-memory par allelism inside a multicore-based compute node, covering both the on-chip parallelism between multiple cores and intra-node parallelism between multiple chips. Second, we will study the issue of work partitioning with respect to heterogeneous multicore ar chitectures, which typically consist of a regular host CPU and a special multicore accelerator. Both the above topics will be related to software development needed for doing parallel finite element computations. Fruitful outcome from this collaborative project, between Center for Biomedial Computing (CBC) and UCSD, will provide CBC with urgently needed computing capability for solving challenging computational problems on modern multicore-based parallel computers. Such an enhanced use of parallelism wil l eventually help promote to the outside world CBC's special finite element software technology.