Fast Multipole Method (FMM) research showcased at the 2018 SIAM PP18 Conference in Tokyo
The University of Bristol’s HPC group has just presented work by one of its PhD students, Patrick Atkinson, on developing an FMM mini-app, called miniFMM. The presentation described how miniFMM has been used to evaluate parallel task frameworks such as OpenMP, Kokkos, TBB, Cilk, OmpSs and BOLT. New results were shown which evaluated the performance of Kokkos’ new GPU-side parallel tasking capability. Performance results were also shown comparing miniFMM performance across GPUs including NVIDIA Volta V100 and Pascal P100, as well as Intel Skylake Xeons and KNL Xeon Phis.
The presentation was part of the SIAM PP18 minisymposium on Hierarchical Low Rank Approximation Methods, at Waseda University in Tokyo, Japan. The slides from the presentation are available online. The miniFMM source code is also freely available in the UoB-HPC Github repo. Please get in touch if you’d like to try the Kokkos GPU tasking version as this is still in experimental form.
Simon McIntosh-Smith, Professor of High Performance Computing, Head of the HPC Group, University of Bristol. Follow @simonmcs for more news from the HPC research group in Bristol.