Re: Working patch for FFTW 3.0.x and Nd FFT's

[David Bateman]

According to John W. Eaton <address@hidden> (on 02/17/04):
> I'm not sure of the best locations for them since I don't know the
> typical sizes of the data.  Should they always go inside the innermost
> loops?

I'd rather they not go in the inner most loop. But the problem is what 
happens in a case like "a = rand(2^20,1); fft(a)" which will call 
convert_packcomplex_1d and will only call the outer loop once, but
the inner one "2^20 / 2 - 1" times. So, now that I think about it, I
really don't know either. It depends really on what data is handed to
them.

The fact is that these two routines occupy a minor proportion of the
time relative to the FFTW code itself. So a compromise might be to
just have a single call to OCTAVE_QUIT at the top of
convert_packcomplex_1d and convert_packcomplex_Nd. 

Regards
David

-- 
David Bateman                                address@hidden
Motorola CRM                                 +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 1 69 35 77 01 (Fax) 
91193 Gif-Sur-Yvette FRANCE

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary