Your test program is primarily written for executing single-threaded FFTs only. And the single-threaded FFT performance of amd-fftw is on par or better.

For multi-threaded FFT tests, you need to modify your test program. Also, slight corrections are required to your compilation command.

Please refer to the below link where your question has been answered in detail: at https://github.com/amd/amd-fftw/issues/19