history.md

History

As a bit of history of this project, it started in 2008 with PyFMM, a 2D serial prototype in Python; then followed PetFMM in 2009, a PETSc-based parallel code with heavy templating [@cruz2011petfmm]; the third effort was GemsFMM in 2010, a serial code with CUDA kernels for execution on GPUs [@yokota2011gems]. Another student in the group felt that exafmm-alpha overused class inheritance and was inadequate for his application, so he and a collaborator re-used only the kernels and implemented another tree construction in fmmtl. The first author of this paper began working with exafmm-alpha in 2017, but at this point the mission was clear: simplify and aim for reusability. That work led to the sixth implementation of FMM in the group, still called exafmm, but it is not what we present here and we haven't published about it. With this new version of ExaFMM (called exafmm-t), we aim to bring the FMM to a broader audience and to increased scientific impact.

Versions

exafmm-t

History: 2017/12/11 - Now
Branch: gpu, vanilla-m2l
Kernel: LaplaceKI, HelmholtzKI, YukawaKI Periodic: no
SIMD: vec.h
Thread: OpenMP loops and OpenMP tasks MPI: none
GPU: P2P, M2L
Build: autoconf Wrapper: none
Plotting: none

exafmm

History: 2017/03/03 - Now
Branch: dev, learning
Kernel: Laplace, LaplaceKI, Helmholtz, Stokes
Periodic: yes
SIMD: vec.h
Thread: OpenMP tasks
MPI: HOT (global histogram sort)
GPU: no
Build: autoconf
Wrapper: none
Plotting: Python

exafmm-alpha > exafmm-beta

History: 2012/07/21 - 2017/03/01
Branch: develop, sc16
Kernel: Laplace, Helmholtz, BiotSavart, Van der Waals
Periodic: yes
SIMD: Agner's vectorclass
Thread: OpenMP, Cilk, TBB, Mthreads
MPI: ORB (bisection, octsection)
GPU: separate code (Bonsai hack)
Build: autoconf & CMake
Wrapper: CHARMM, GROMACS, general MD, PetIGA
Plotting: none

exafmm-alpha/old + vortex_method > exafmm-alpha

History: 2010/12/22 - 2012/07/21
Branch: none
Kernel: Laplace, Van der Waals, Biot Savart, Stretching, Gaussian
Periodic: yes
SIMD: none
Thread: QUARK
MPI: ORB (global nth_element)
GPU: offload all kernels
Build: Makefile
Wrapper: MR3 compatible MD
Plotting: VTK

old_fmm_bem

History: mid 2010 - late 2010
Branch: none
Kernel: Laplace, Laplace Gn, Helmholtz
Periodic: no
SIMD: none
Thread: none
MPI: Allgather LET
GPU: offload all kernels
Build: Makefile
Wrapper: none
Plotting: none

old_fmm_vortex

History: early 2010 - mid 2010
Branch: none
Kernel: Laplace, Biot Savart, Stretching (transpose,mixed), Gaussian
Periodic: yes (shear, channel)
SIMD: none
Thread: none
MPI: Allgather LET
GPU: offload all kernels
Build: Makefile
Wrapper: none
Plotting: none

Other FMM codes

ASKIT
Bosnsai
ChaNGa
FDPS
KIFMM
KIFMM new
Modylas
PKIFMM
PEPC
pfalcON
PVFMM
Salmon treecode
ScaFaCos
ScalFMM