[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Working patch for FFTW 3.0.x and Nd FFT's
From: |
David Bateman |
Subject: |
Re: Working patch for FFTW 3.0.x and Nd FFT's |
Date: |
Tue, 17 Feb 2004 09:53:09 +0100 |
User-agent: |
Mutt/1.4.1i |
According to John W. Eaton <address@hidden> (on 02/16/04):
> On 16-Feb-2004, David Bateman <address@hidden> wrote:
>
> | Given all of the testing I've given the code, I now consider this
> | patch ready for inclusion into 2.1.54....
>
> OK, I would like to apply the patch, but after a quick look at the
> code, I have a couple of comments and questions.
>
> In several places, you are using const_cast to cast away const. Why
> is this necessary? Does the FFTW code really modify its argument, or
> is it simply missing const qualifiers. If the former, then I think we
> need to find a way to avoid requiring the cast, otherwise this could
> do the wrong thing when mixed with Octave's reference counting. If it
> is the latter, then I think we should ask the FFTW maintainers to add
> the const qualifiers.
When FFTW3 creates plans with anything but the FFTW_ESTIMATE flag, then it
using the *in and *out arrays as testing space for the plans, effectively
destorying the data in them. However with FFTW_ESTIMATE doesn't touch them
except to check data alignment.
So const_cast is necessary to force FFTW to accept the argument. This will
only cause a problem we change from FFTW_ESTIMATE. I see no reason for this
as I've also supplied a function fft_wisdom to create wisdom seperately from
the fft itself.
> Thanks for being careful to include OCTAVE_QUIT in most loops so that
> Octave can be interrupted when these calculations are happening. I
> did notice that the function convert_packcomplex_1d and
> convert_packcomplex_Nd contain loops and do not use OCTAVE_QUIT. Is
> that intentional? Is that code so simple that it can never run for a
> long time, even if a user accidentally passes in a very large array?
That is more of an oversight. Feel free to add the OCTAVE_QUITs there
Cheers
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
- Working patch for FFTW 3.0.x and Nd FFT's, David Bateman, 2004/02/16
- Working patch for FFTW 3.0.x and Nd FFT's, John W. Eaton, 2004/02/16
- Re: Working patch for FFTW 3.0.x and Nd FFT's,
David Bateman <=
- Re: Working patch for FFTW 3.0.x and Nd FFT's, John W. Eaton, 2004/02/17
- Re: Working patch for FFTW 3.0.x and Nd FFT's, David Bateman, 2004/02/17
- Re: Working patch for FFTW 3.0.x and Nd FFT's, John W. Eaton, 2004/02/17
- Re: Working patch for FFTW 3.0.x and Nd FFT's, Paul Kienzle, 2004/02/17
- Re: Working patch for FFTW 3.0.x and Nd FFT's, John W. Eaton, 2004/02/18
- Re: Working patch for FFTW 3.0.x and Nd FFT's, Dmitri A. Sergatskov, 2004/02/18
- Re: Working patch for FFTW 3.0.x and Nd FFT's, John W. Eaton, 2004/02/18
- Re: Working patch for FFTW 3.0.x and Nd FFT's, Dmitri A. Sergatskov, 2004/02/18
- Re: Working patch for FFTW 3.0.x and Nd FFT's, Paul Kienzle, 2004/02/18
- Re: Working patch for FFTW 3.0.x and Nd FFT's, John W. Eaton, 2004/02/18