matrixberry / efftw

Modern C++ FFTW Wrapper for Eigen library

Eigen-FFTW

This repository is brand new and under heavy development!

E-FFTW is a modern C++20 wrapper library around FFTW for Eigen.

• Supports 1D and 2D FFTs
• Single header file: #include "efftw/efftw.hpp"

Including

include(FetchContent)

FetchContent_Declare(
  efftw
  GIT_REPOSITORY https://github.com/stephenberry/efftw
  GIT_TAG main
  GIT_SHALLOW TRUE
)
FetchContent_MakeAvailable(efftw)

target_link_libraries(${PROJECT_NAME} PRIVATE efftw::efftw)

Note: FFTW and Eigen dependencies are not included or installed by linking to efftw::efftw

Example

#include "efftw/efftw.hpp"

int main()
{
   // init_threads must be called before the library is used
   efftw::init_threads(4); // set FFTW to use 4 threads

   // build a complex matrix
   Eigen::MatrixXcd mat(N, N);

   std::mt19937_64 g{}; // random number generator
   std::uniform_real_distribution<double> dist{0.0, 1.0};

   for (size_t r = 0; r < N; ++r) {
      for (size_t c = 0; c < N; ++c) {
         mat(r, c) = { dist(g), dist(g) };
      }
   }
   
   efftw::f2 fft{mat}; // FFTW planning on construction (may be reused)
   fft(); // compute the FFT of mat in place
}

API

using namespace efftw;
// classes
f1{vec} // 1D forward FFT (not normalized)
f2{mat} // 2D forward FFT
i1{vec} // 1D inverse FFT (1/(rows) normalization)
i2{mat} // 2D inverse FFT (1/(rows * cols) normalization)

// functions
shift1(vec) // 1D forward FFT shift
shift2(mat) // 2D forward FFT shift
inv_shift1(vec) // 1D inverse FFT shift
inv_shift2(mat) // 2D inverse FFT shift

Important!

E-FFTW classes take references to Eigen types. Do not delete the matrix or resize it without rebuilding the EFFTW class.

The E-FFTW classes maintain the FFTW plan, which is deleted in the E-FFTW class destructors. The classes are used to keep the plan alive and allow the same matrix memory to be used multiple times. It is inefficient to rebuild the plan, but the plan needs to be rebuilt if the size of the matrix or vector changes.

Alias Type Deduction

// For clang, alias type deduction is not yet supported, so you will need to write:
efftw::f2<decltype(mat)> fft{mat};