git clone https://github.com/vmarkovtsev/FFTF.git

./autogen.sh build --enable-gpl --enable-opencl --enable-cuda --prefix=/opt
cd build && make -j$(getconf _NPROCESSORS_ONLN)
sudo make install

#include <assert.h>  // for assert()
#include <fftf/api.h>
#include <stdlib.h>  // for atexit()

// Scan for libraries
FFTFBackendbackends = fftf_available_backends(NULL, NULL);

// Check if FFTW3 was successfully loaded
assert(backends[FFTF_BACKEND_FFTW3].path != NULL);

// Give it maximal priority
fftf_set_backend_priority(FFTF_BACKEND_FFTW3,
                          BACKEND_MAX_INITIAL_PRIORITY + 1);

// Note: the alternative way to use that backend is just
assert(fftf_set_backend(FFTF_BACKEND_FFTW3) == FFTF_SET_BACKEND_SUCCESS);
// without even scanning for available backends

// Schedule the FFTF resources cleanup
atexit(fftf_cleanup);

// Transform N complex numbers
const int N = 2048;
floatin = fftf_malloc(N2sizeof(float));
fill_with_numbers(in, N);

// Prepare the space for the result
floatout = fftf_malloc(N2sizeof(float));

// Obtain the calculation plan (instance)
FFTFInstance instance = fftf_init(FFTF_TYPE_COMPLEX, FFTF_DIRECTION_FORWARD,
                                 FFTF_DIMENSION_1D, N, FFTF_NO_OPTIONS,
                                 in, out);

// Calculate the forward 1-dimensional complex FFT
fftf_calc(instance);

// Free the plan
fftf_destroy(instance);

// Free in and out arrays
fftf_free(in);
fftf_free(out);