How to setup Octave for GPU CUDA computing usind FFT-routines

[Mattias Trimpop]

Hello!

I spend a lot of time for using CUDA-FFT's with Octave without success.
Perhaps anyone can help me?

I am using Octave (on Ubuntu 16.04 an a NVidia-Graphics card with 2 GB 
Ram) to calculate two-dimensional (large) fields. That uses a lot of 
FFTs and IFFTs. To fasten the calculation I tried the CUDA-libraries. 
cuBLAS (from CUDA V8.0) is working fine but has no optimisation for the 
used FFT-Routines. The video under "Drop-in Acceleration on GPUs with 
Libraries" - "Learning Libraries" shows how to use the CUDA-FFT instead 
of FFTW. I managed to install the octave sources (V. 4.0.3) and to 
compile a new binary that workes well. Unfortunately I am not so 
familiar with coding under Linux that I managed to replace the FFTW with 
the CUFFT.
Has anyone made this and can provide me with the changed sources/files 
or a "howto" (perhaps with another octave-Version?)? That would be very 
helpful!

Thanks in advance!

Mattias Trimpop