Install fftw with Homebrew, apk, apt, dnf, MacPorts, Nix, pacman

Project links

• Homepage
• Repository
• Upstream docs
• Package manager page

Commands and aliases

• fftw-wisdom
• fftw-wisdom-to-conf
• fftwf-wisdom
• fftwl-wisdom

Project history

FFTW was developed at MIT by Matteo Frigo and Steven G. Johnson. The official site frames its goal as a portable free FFT library that can compete with vendor-tuned codes while remaining usable across architectures without application changes.

The early public release line began with FFTW 1.0 in March 1997. The 1998 ICASSP paper described FFTW as an adaptive software architecture for FFTs, and the 1999 PLDI paper described the codelet generator behind its fast generated kernels.

FFTW 3.0 arrived in April 2003 with an API incompatible with FFTW 2.x, trading compatibility for performance and generality. The current manual documents the 3.x planner model: create a plan once, reuse it for repeated transforms, and optionally save planning knowledge as wisdom.

The FFTW3 design paper appeared in Proceedings of the IEEE in 2005 and remains the preferred general reference named by the project. Later 3.3 releases added modern CPU and parallel features, including AVX, ARM Neon, MPI, Fortran 2003 interfaces, AVX2, AVX512, VSX, SVE, LoongArch SIMD, and additional cycle counters.

Adoption history

FFTW is distributed by mainstream Unix package managers, including Homebrew, Debian/Ubuntu, Fedora, Arch, MacPorts, Nix, and Alpine according to the package facts in this batch. Its availability in both operating-system packages and source builds reflects its role as a common dependency for scientific, signal-processing, audio, imaging, and simulation software.

The project also maintains benchFFT results and positions FFTW against other public FFT implementations and vendor libraries. That benchmark culture helped make FFTW a default comparison point for package maintainers deciding whether a package should use a system FFT library or a bundled implementation.

How it is used

Library users link against FFTW and create plans for one-dimensional, multidimensional, real, complex, DCT/DST, or Hartley transforms. The manual describes basic, advanced, and guru interfaces, with the basic interface intended for most users and guru interfaces for custom layouts and strides.

The Homebrew package also installs command-line utilities such as `fftw-wisdom`, `fftw-wisdom-to-conf`, and precision-specific wisdom helpers. These are package-visible tools around FFTW's planning cache rather than the main reason most software depends on the library.

Why package nerds care

FFTW matters to package nerds because it sits at the intersection of ABI policy, CPU feature selection, license policy, and performance tuning. Distribution maintainers must choose which SIMD variants to enable, whether to build thread and MPI variants, and how to expose the GPL library to downstream packages.

It is also a canonical example of a library whose best behavior depends on installation choices and local hardware. The upstream release notes even warn distribution maintainers that enabling every SIMD option can increase planning time for little general-purpose benefit.

Timeline

• 1997: FFTW 1.0 released.
• 1998: FFTW 2.0 released and the ICASSP adaptive-architecture paper published.
• 1999: PLDI paper described FFTW's FFT compiler and codelet generator.
• 2003: FFTW 3.0 released with an incompatible, more general API.
• 2005: FFTW3 design and implementation paper published in Proceedings of the IEEE.
• 2011: FFTW 3.3 added AVX, MPI distributed-memory transforms, and a Fortran 2003 API.
• 2026: FFTW 3.3.11 added SVE, LoongArch SIMD, and new cycle-counter support.

Related projects

• FFTW is commonly compared with vendor FFT libraries and public FFT implementations through benchFFT. It is also related to downstream numerical packages that choose between bundled FFT code and a system FFTW dependency.

Sources

• https://www.fftw.org/
• https://www.fftw.org/doc/Introduction.html
• https://www.fftw.org/release-notes.html
• https://www.fftw.org/fftw-paper-ieee.pdf
• https://github.com/FFTW/fftw3
• https://formulae.brew.sh/formula/fftw