The parallel FFTW3 distributes different portions of a large data to different processors, so that each processor obtains and manipulates only a small fraction of the large size data that wants to be fourier transformed.
I am wondering, at the end of the transformation, as the transformed data are locating at different processors, is there any built-in routine in FFTW3 to gather all these small portions from different processors to recover the entire data set, for example, at the root processor. Hope this question is clear to you guys and I am looking forward to your suggestions.