Flatiron Institute Nonuniform Fast Fourier Transform library: FINUFFT

Principal author Alex H. Barnett ([email protected]), main co-developers Jeremy F. Magland, Ludvig af Klinteberg, Yu-hsuan "Melody" Shih, Libin Lu, Joakim Andén, Marco Barbone, Robert Blackwell, and Martin Reinecke; see docs/ackn.rst for full list of contributors. ​

This is a lightweight CPU library to compute the three standard types of nonuniform FFT to a specified precision, in one, two, or three dimensions. It is written in C++ with interfaces to C, Fortran, MATLAB/octave, Python, and (in a separate repository) Julia. It now also integrates the GPU CUDA library cuFINUFFT.

Docs and installation

Please see the online documentation which can also be downloaded as a PDF manual, and a project overview. You will also want to see CPU example codes in the directories examples, test, fortran, matlab/test, matlab/examples, python/finufft/test, etc, and GPU examples in examples/cuda, test/cuda, etc.

If you cannot build via cMake, try the makefile. Python users try pip install finufft. See the docs for details. See our GitHub Issues for tips.

Supported CPU platforms

The matrix below is mirrored from the CI workflows (.github/workflows/cmake_ci.yml, .github/workflows/powerpc.yml); status cells are live badges. To regenerate after editing a workflow, run python tools/gen_platform_table.py and paste the output back into this section.

Platform	Toolchains (per-job CI status)
MSYS2 MINGW64
macos-14
macos-15
macos-15-intel
manylinux_2_28
powerpc64
powerpc64le
ubuntu-22.04
ubuntu-24.04
windows-2022
windows-2025

Each badge labels its toolchain and colours green/red with the latest status on master. CMake-CI rows on Linux/macOS exercise both DUCC FFT and FFTW; Windows builds DUCC FFT only. (make) rows share their workflow's overall status (legacy GNU-make build path); PowerPC rows likewise share powerpc.yml's overall status (cross-compile + QEMU).

See also the Supported CPU platforms section of the rendered install docs.

If you prefer to read text files, the source to generate the above documentation is in human-readable (mostly .rst) files as follows:

• docs/install.rst : installation and compilation instructions
• docs/install_gpu.rst : installation and compilation for GPU (CUDA)
• docs/dirs.rst : explanation of directories and files in the package
• docs/math.rst : mathematical definitions
• docs/cex.rst : example usage from C++/C
• docs/c.rst : documentation of C++/C function API
• docs/c_gpu.rst : documentation of C++/C function API for GPU library
• docs/opts.rst : optional parameters
• docs/error.rst : error codes
• docs/trouble.rst : troubleshooting and accuracy advice
• docs/performance.rst : CPU timing benchmarks compared to older versions
• docs/tut.rst and tutorial/* : tutorial application examples
• docs/fortran.rst : usage examples from Fortran, documentation of interface
• docs/matlab.rst and docs/matlabhelp.raw : using the MATLAB/Octave interface
• docs/python.rst and python/*/_interfaces.py : using the Python interface
• docs/python_gpu.rst : Python interface to GPU library
• docs/julia.rst : information for Julia users
• docs/nfft_migr.rst and docs/cufinufft_migration : guides for migration
• docs/impl_gpu.rst: implementation notes for GPU code
• docs/devnotes.rst: notes/guide for developers
• docs/related.rst : other recommended NUFFT packages
• docs/users.rst : some known users of FINUFFT, dependent packages
• docs/ackn.rst : authors and acknowledgments
• docs/refs.rst : journal article references (ours and others)

License

For license info for FINUFFT (CPU/GPU source), see LICENSE. For info for dependencies and contributed code, see NOTICE.

Support

FINUFFT/cuFINUFFT was developed at, and is supported by, the Flatiron Institute at the Simons Foundation.

Citing

If you find (cu)FINUFFT useful in your work, please star this repository and cite it and the following. It will help us to improve the library if you also describe your use case parameters here.

For FINUFFT (CPU library):

A parallel non-uniform fast Fourier transform library based on an ``exponential of semicircle'' kernel. A. H. Barnett, J. F. Magland, and L. af Klinteberg. SIAM J. Sci. Comput. 41(5), C479-C504 (2019).

For cuFINUFFT (GPU library):

cuFINUFFT: a load-balanced GPU library for general-purpose nonuniform FFTs, Yu-hsuan Shih, Garrett Wright, Joakim Andén, Johannes Blaschke, Alex H. Barnett, PDSEC2021 workshop of the IPDPS2021 conference. https://arxiv.org/abs/2102.08463