Hello,
Just wondering, I realise we're still in mayer_fft time, and that there is an FFTW patch out there. However, I remember Miller saying he was using Ooura, but I can't see this being committed to CVS at all.
Some of us are trying to use FFTW instead of mayer, but finding that initialisation times can be a problem, and that we could rather use an FFT library that is more DSP chain friendly. I assume Miller's choice of Ooura has to do with that? (more speed than mayer, faithful to dsp chain?)
Is anyone else out there using Ooura fft? Any idea of timeline for it to make it into the sources?
David
On Tue, 25 Sep 2007, David Plans Casal wrote:
Some of us are trying to use FFTW instead of mayer, but finding that initialisation times can be a problem,
Perhaps if you shared fftw_plan across all objects that use the same block size, you'd save clicks? Perhaps that just loading the fftw class that you are using could prepare plans for several common block sizes? For example every power-of-two from 64 to 4096.
and that we could rather use an FFT library that is more DSP chain friendly. I assume Miller's choice of Ooura has to do with that?
I doubt that Miller tried FFTW3...?
_ _ __ ___ _____ ________ _____________ _____________________ ... | Mathieu Bouchard - tél:+1.514.383.3801, Montréal QC Canada
Hello,
On 28 Sep 2007, at 05:57, Mathieu Bouchard wrote:
Perhaps if you shared fftw_plan across all objects that use the same block size, you'd save clicks? Perhaps that just loading the fftw class that you are using could prepare plans for several common block sizes? For example every power-of-two from 64 to 4096.
Ok that sounds like the right magic; thank you!
and that we could rather use an FFT library that is more DSP chain friendly. I assume Miller's choice of Ooura has to do with that?
I doubt that Miller tried FFTW3...?
Miller?
David
-
David Plans Casal -
Mathieu Bouchard