A fork of ExaFMM-T

This is fork of the ExaFMM-t project and very much a work in progress. So what are the big changes so far? The good:

• Improved comments and code clarity.
• Easier to compile everywhere. Developed with MSVC, but intended to compile everywhere at some point.
• Eigen is now used for linear algebra.
  • No longer dependent on BLAS or LAPACK libaries.
  • Everything is header-only.
  • This makes the code easier to read.
• No more virtual. Theoretically improved performance.
• All AMD64 intrinsics have been removed.
  • Should compile on non-AMD64 architectures.
• Heavier use of C++17:
  • All types are now handled using templates - no macros required.
  • Templated sections of the code no longer need type specific specializations.
  • Constexpr allows more to be done at compile time:
    • Some array sizes are now known at compile time.
    • Some pre-computation can occur at compile time.

The bad:

• The python interface is gone.
• This code still needs to be validated.

To do:

• Change interface for ease of use.
• Improved testing.
• Template based on FMM order.
• Octree representation
  • Source leaves, target leaves, source nodes and target nodes should all be different types.
• Regularised particles.
• Vector potentials.
• GPU acceleration:
  • SYCL-2020 (for this, the codebase must stay as C++2017).
• Non-particle sources.

What was exafmm-t

The original exafmm-t is available here and no longer appears to be actively developed.

"exafmm: a high-performance fast multipole method library with C++ and Python interfaces", Tingyu Wang, Rio Yokota, Lorena A. Barba. The Journal of Open Source Software, 6(61):3145 (2021). doi:10.21105/joss.03145

exafmm-t is a kernel-independent fast multipole method library for solving N-body problems. It provides both C++ and Python APIs. We use pybind11 to create Python bindings from C++ code. exafmm-t aims to deliver compelling performance with a simple code design and a user-friendly interface. It currently supports both potential and force calculation of Laplace, low-frequency Helmholtz and modified Helmholtz (Yukawa) kernel in 3D. In addition, users can easily add other non-oscillatory kernels under exafmm-t's framework.

The full documentation is available here.