Re: [linux-audio-dev] Re: [Jackit-devel] Re: [linux-audio-user] [ANNOUNCE] jack_convolve-0.0.1

From: Steve Harris <S.W.Harris@email-addr-hidden>
Date: Sat Jan 29 2005 - 21:44:47 EET

On Sat, Jan 29, 2005 at 07:20:39 +0100, Florian Schmidt wrote:
> On Sat, 29 Jan 2005 06:55:31 +0100
> Florian Schmidt <mista.tapas@email-addr-hidden> wrote:
>
> > tarball[s] here [minor optimization attempts since last tarball]:
> >
> > http://affenbande.org/~tapas/jack_convolve/
>
> about optimizations: like Anders Torger notes in his docs to brutefir:
>
> http://www.ludd.luth.se/~torger/brutefir.html#bruteconv_4
>
> the cycle burner is the complex multiplication. So if anyone has SIMD
> assembler routines for complex multiplication of two float buffers
> (fftwf_complex *) in his cupboard, please bring it out :)
>
> jack_convolve uses complex data format, not half-complex like brutefir.
> so i suppose i can't simply rip his code ;)

I'd recommend the half-complex format, it makes hte FFT much faster and
reduces the ammount of memory you have to touch. The only pain is that the
format is a bit wierd to work with.

- Steve
Received on Sun Jan 30 00:15:44 2005

This archive was generated by hypermail 2.1.8 : Sun Jan 30 2005 - 00:15:45 EET