High Scalability Codes for the Fast Multipole Method

Paralleling the fast multipole method in electromagnetics for the solution of large-scale problems using supercomputers is addressed here. Paralleling is achieved using different approaches. For distributed memory computers, message passing is the best selection to improve performance, while threads-based fork-and-join paralleling approaches are used in shared memory computers. Mixed computers may require a hybrid methodology, combining the MPI standard for message passing and POSIX threads or OpenMP pragmas to create low-level threads. A hybrid MPI/OpenMP programing is currently being implemented to obtain high efficiency codes in mixed computers.