tmolteno / template-fft

Fast Header only C++ FFT and Inverse FFT using templates

Template-FFT

Fast Header only C++ FFT and Inverse FFT using templates.

Here is the source code for a complex FFT algorithm that uses template metaprogramming. It is quite fast (nearly as fast as fftw) and can be specialized to any type. We will be creating a fixed-point friendly FFT as well in the near future.

Using the C++ template

Create the file below, and compile it with your favourite C++ compiler. Requires Eigen for handling vectors and matrices.

#include "fft_complex.h"
#include <Eigen/Dense>
using namespace std;

#define LOG_LENGTH 13
#define N (1<<LOG_LENGTH)

typedef complex<double> fft_complex;

#include <iostream>

int main()
{
    CFFT<N,double> cfft;
    Matrix<complex<double>, N, 1> cdata;
    cdata.setRandom();

    std::cout << "Perform 1000 FFT and INVERSE FFT " << endl;
    std::cout << "FFT Length= " << N << endl;
    
    for (int i=0;i<1000;i++) {
        cfft.fft(cdata);
        cfft.ifft(cdata);
    }
return 0;
}

Copyright

Author: Tim Molteno, [email protected]

Based on the article "A Simple and Efficient FFT Implementation in C++" by Volodymyr Myrnyy with an inverse FFT modification.

Licensed under the GPL v3.

template-fft's Issues

Adding constraints.

I thought I'd contribute a tiny snippet of code to constrain the integer values used for template specialization to powers of two using the new c++20 requires feature.

#include <concepts>

constexpr bool is_power_of_two(std::integral auto value) {
    return (value > 0) && ((value & (value - 1)) == 0);
}

template<std::integral auto i>
requires (is_power_of_two(i))
struct Foo{

};