Re: How to setup Octave for GPU CUDA computing

[Sergei Steshenko]

>________________________________
> From: Andreas Weber <address@hidden>
>To: address@hidden 
>Sent: Saturday, July 9, 2016 4:40 PM
>Subject: Re: How to setup Octave for GPU CUDA computing
> 
>
>Am 09.07.2016 um 11:16 schrieb Zoltán Szabó:
>>
>> I am using the xcorr
>> <https://sourceforge.net/p/octave/signal/ci/stable/tree/inst/xcorr.m> 
>> function
>> from Octave's signal package on Linux. It has a bunch of fft, ifft,
>> conj, etc in the script. How to configure/compile Octave to use CUDA
>> capabilities specifically for this xcorr function? Maybe using cuFFT
>> somehow? Which gpu-accelerated library is the best for this function?
>>
>
>Crosspost on stackoverflow:
>http://stackoverflow.com/questions/38277024/how-to-setup-octave-for-gpu-cuda-computing
>
>This has been discussed many times and AFAIK the consents was that
>linking GNU Octave to the nonfree cuBlas library violates the GPL
>although you find some tutorials
>(https://devblogs.nvidia.com/parallelforall/drop-in-acceleration-gnu-octave/)
>
>-- Andy
>
>



Nothing violates GPL if the linked code is not being distributed (see the GPL 
FAQ).

Furthermore, copying code from one computer to another inside the same 
organization is NOT considered to code distribution according to the GPL FAQ.

So, "just do it". I.e. replace FFTW with CUDA equivalent, and BLAS with cuBLAS 
and use it in the comfort of yor home or organization.

--Sergei.