better fft performance options

[Stefan Sauer]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 09/19/2015 05:52 PM, Sebastian Dröge wrote:
> On Fr, 2015-09-18 at 07:18 -0700, John Ludwig wrote:
>> i’m developing a gstreamer element which makes extensive use of FFTs
>> on audio signals.  i am CPU bound currently and am dropping a lot of
>> frames, I am wondering if anyone has done any work to look at
>> hardware-based improvements to the gstreamer FFT routines, for
>> instance using Intel’s IPP lib, or using a GPU based implementation.
>>  I didn’t see anything on a  quick scan thru the repo, but I feel
>> like someone must have already done some work in this area.
>>  Appreciate any pointers…thanks
>
> Our fft library is based on kissfft, which is just that. Very small and
> simple implementation in C, not handcrafted assembly.
> http://kissfft.sourceforge.net/
>
> If you need something very fast, people are usually using fftw.
Btw. we chose kissfft since it has more liberal licensing. fftw is GPL.
It would be wonderful to have the butterfly op orc optimized.

Stefan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
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=0p7e
-----END PGP SIGNATURE-----