[Sc-devel] FFT ugen using FFTW

[James McCartney]

The FFT ugen is not currently using vDSP. There is some preliminary  
vDSP code, but it is not turned on, IIRC. It is using the Green  
implementation.

For nonrealtime FFT I can see having nonpower of two. But for overlap  
adding of real time streams, which is what the FFT ugen does, I'm not  
sure why it would be needed.

For Linux, FFTW is probably the best solution.

--- james mccartney @ iphone

On Jul 31, 2007, at 5:23 AM, "Dan Stowell" <danstowell at gmail.com> wrote:

> Thanks for the pointer, James.
>
> So, two questions:
> (1) How is the relative performance of FFT ugen vs FFTW ugen on Linux
> / Windows? I hope people can feed back.
> (2) Is there any strong motivation to enable non-power-of-two FFT
> sizes? If yes, then maybe we stick with FFTW. If no, maybe we use vDSP
> on mac and FFTW on non-mac. (Other options exist, of course... discuss
> away...)
>
> Dan
>
>
> 2007/7/31, James McCartney <asynth at gmail.com>:
>> On the Mac vDSP should be used instead of FFTW. By FFTW's own
>> benchmarks vDSP wallops FFTW on PPC for single precision real data
>> power of two transforms.
>>
>> http://www.fftw.org/speed/G4-1.06GHz-macosx/ibook-macosx.1d.srxx.p2.png
>> http://www.fftw.org/speed/G5-2GHz-macosx/g5-macosx.1d.srxx.p2.png
>>
>> Unfortunately there are no benchmarks for FFTW vs vDSP on Intel Macs
>> but I believe that FFTW gets a similar ass kicking there.
>>
>>
>> On 7/30/07, Dan Stowell <danstowell at gmail.com> wrote:
>>
>>>
>>> PPC G4 Mac PowerBook, 1.67 GHz:
>>>        Idle:             2.9 %
>>>        FFT:    48.5 %
>>>        FFTW:   50.1 %
>>>
>>> Intel Core 2 Duo iMac, 2 GHz:
>>>        Idle:             2.6 %
>>>        FFT:    31.5 %
>>>        FFTW:   23.2 %
>>>
>>> I'd appreciate reports of tests on other systems, and other feedback
>>> too of course.
>>>
>>> Dan
>>>
>>> --
>>> http://www.mcld.co.uk
>>>
>>> _______________________________________________
>>> Sc-devel mailing list
>>> Sc-devel at create.ucsb.edu
>>> http://www.create.ucsb.edu/mailman/listinfo/sc-devel
>>>
>>>
>>>
>>
>>
>> --
>> --- james mccartney
>> _______________________________________________
>> Sc-devel mailing list
>> Sc-devel at create.ucsb.edu
>> http://www.create.ucsb.edu/mailman/listinfo/sc-devel
>>
>
>
> -- 
> http://www.mcld.co.uk
> _______________________________________________
> Sc-devel mailing list
> Sc-devel at create.ucsb.edu
> http://www.create.ucsb.edu/mailman/listinfo/sc-devel