ExaFMM: An open source library for Fast Multipole Methods aimed towards Exascale systems

The figure shows tree cases: the ones on th eleft and right satisfy the multipole acceptance criterion (MAC), while the one in the center does not and is pushed to the stack. Above — The flow of the vortex method caluclation using parallel FMM, where the domain is partitioned by an orthogonal recursive bisection and communication of the local essential tree (LET) is overlapped with the GPU calculation of the local tree. The remaning parts of the LET are calculated and updates are performed locally.