Quick start

The following code is a simplified version of 1d_c2c_single_example.cpp available from NVPLSamples, showing how to use NVPL FFT.

#include <iostream>
#include <vector>
#include <complex>
#include <nvpl_fftw.h>

int main(int argc, char *argv[]) {
    const int n = 4;
    const int howmany = 2;

    std::vector<std::complex<float>> in(howmany * n);
    std::vector<std::complex<float>> out(howmany * n);

    fftwf_complex *in_data  = reinterpret_cast<fftwf_complex*>(in.data() );
    fftwf_complex *out_data = reinterpret_cast<fftwf_complex*>(out.data());

    fftwf_plan plan = fftwf_plan_dft_1d(n, in_data, out_data, FFTW_FORWARD, FFTW_ESTIMATE);

    int i = 0;
    for (auto& v: in) {
        v = {(float)i, -(float)i};
        i++;
    }

    for(int b = 0; b < howmany; b++) {
        fftwf_execute_dft(plan, in_data+b*n, out_data+b*n);
    }

    for (const auto& v: out) {
        std::cout << v << "\n";
    }

    fftwf_destroy_plan(plan);
    fftwf_cleanup();
    return 0;
}

This example computes howmany batches of a one-dimensional forward FFT with size n. It does the following:

1. Allocates the input and output data

2. Creates a 1D FFTW plan with specified size n and transform direction FFTW_FORWARD

3. Initializes the input data.

4. Executes the plan. One batch at a time.

5. Outputs the results.

6. Destroys the plan.

The example can then be compiled and run as follows:

$ g++ app.cpp -I/path/to/nvpl_fft/include -L/path/to/nvpl_fft/lib -lnvpl_fftw -o app
$ ./app
(6,-6)
(0,4)
(-2,2)
(-4,0)
(22,-22)
(0,4)
(-2,2)
(-4,0)

Tip

For an application that dynamically links to the FFTW3 libraries, one could test NVPL FFT by running the application with LD_PRELOAD=/path/to/nvpl_fft/lib/libnvpl_fftw.so.