J.M.Taboada, L.Landesa, J.Rivero, G.Gajardo-Silva, M.Amaya, F.Obelleiro, J.L.Rodríguez, M.G.Araújo, J.M.Bértolo, J.C.Mouriño, A.Gómez, C.Gómez, J.L.González-Sánchez: «Supercomputing aware electromagnetics«.VII Encuentro Ibérico de Electromagnetismo Computacional. 20-23. Monfragüe, Cáceres. Mayo de 2010.
Abstract
The Fast Fourier Transform (FFT) extension of the conventional Fast Multipole Method (FMM) has demonstrated
that it reduces the matrix vector product (MVP) complexity while preserving the propensity for parallel scaling of the single level
FMM. An efficient hybrid MPI/OpenMP parallel implementation of the FMM-FFT and, subsequently, an improved nested scheme
of the algorithm, and a combination with MLMFMA have been employed successfully for the solution of challenging problems
with hundreds of millions of unknowns.
Index Terms—Fast Multipole Method, Fast Fourier Transform, Multilevel Fast Multipole Algorithm, Supercomputing