|
|
Blitz Support : |
From: Navneet Dalal (Navneet.Dalal_at_[hidden])
Date: 2003-11-27 11:26:19
Hi Julian,
It is a very good idea to collect user submissions.
I cannot commit myself to FFT package (as currently, I am not using FFT
and may not use it much in future also), but I have some other algorithms
and small code which I feel can be used by blitz community in general. I
will make sure that I organize each of these algorithms, document, test
and submit it to blitz sourceforge website in next few weeks.
Coming back to FFT, I agree with Einar that one should write a blitz
interface to FFTW. Writing a general bug free, reliable library for FFTW
is time consuming and big effort. Moreover, FFTW is a highly optimized for
speed.
Related to license, even though blitz may not provided FFTW directly,
users may be asked to download it themselves. And there exists also a
possibility of purchasing non-free licenses of FFTW from MIT.
Besides FFTW, another idea would be to use a template based FFT (much in
spirit of blitz).
I found a c++ file for doing FFT using templates (written by Todd in 1996)
and is present on blitz website
http://www.oonumerics.org/blitz/examples/fft.html
However, I have serious doubts that using all template based FFT code may
be too slow to compile (and thus develop) for a project using FFT heavily.
regards
navneet
PS: Julian, I really appreciate your effort and commitment to blitz (Not
to forget such a wonderful work by Todd).
On Wed, 26 Nov 2003 13:16:54 -0800, Julian C. Cummings
<cummings_at_[hidden]> wrote:
> Navneet,
>
> Your request is very timely, as I am just now getting serious about
> gathering up user submissions for blitz algorithms and adding them to the
> blitz distribution. Another blitz user submitted some code (based on
> Numerical Recipes, I think) for doing FFT's, along with other types of
> transforms and linear algebra routines. I have been playing with this
> submission for a little while, trying to debug some parts of it and
> clean up
> the interface.
> It is a bit overly general because the code author was attempting to
> target
> multiple array packages, but this can be changed. Also, the routines are
> limited to radix-2, do not attempt to distinguish between real or complex
> transforms, and assume that you are transforming along the rows of the
> given
> Array. I think there needs to be additional support added for
> automatically
> transforming along any Array dimension.
>
> If you have a nice interface in mind for how to apply FFT's to blitz
> Arrays,
> I encourage you to submit it as a Request for Enhancement (RFE) on the
> blitz
> sourceforge website. I can probably modify the previously submitted FFT
> routines to match whatever interface you feel is reasonable, and then we
> can
> add support for real FFTs, non-radix-2 transforms, etc.
> Also, if there is a better starting point than Numerical Recipes for the
> basic code, please make a suggestion or submit some code.
>
> Regards, Julian C.
>
> P.S. I did something like this for the Pooma package in the past. In
> that
> case, I found it sensible to cast the FFT support in the form of a class
> in
> which the type (real or complex) and the dimensionality were template
> parameters. The class provided data members to store the wavetable (exp
> or
> sin/cos factors) and temporary workspace needed, with an initialize
> function
> to set these up. It also provided a transform method with parameters to
> indicate the direction of the transform (forward or inverse) and the
> dimension to transform. This might be an appropriate model for FFTs of
> blitz Arrays as well.
>
>
>> -----Original Message-----
>> From: blitz-support-bounces_at_[hidden]
>> [mailto:blitz-support-bounces_at_[hidden]] On Behalf Of
>> Navneet Dalal
>> Sent: Tuesday, November 25, 2003 8:26 AM
>> To: Support list for Blitz++
>> Subject: [Blitz-support] FFT using blitz
>>
>>
>> Hi everybody,
>>
>> I need to use FFT for (2D) blitz's arrays. So I would like to know if
>> someone has implement FFT in blitz arrays before -- or I am
>> on my own?
>>
>> thanks in advance...
>> navneet
>>
>> --
>> Navneet DALAL http://www.inrialpes.fr/lear/people/Dalal
>>
>> INRIA Rhone-Alpes Ph(work): +33 (0)476 61 54 97
>> ZIRST-655, av. de l'Europe - 38334 FAX : +33 (0)476 61 54 54
>> Montbonnot St. Martin, France Mobile : +33 (0)677 54 33 33
>> _______________________________________________
>> Blitz-support mailing list
>> Blitz-support_at_[hidden]
>> http://www.oonumerics.org/mailman/listinfo.cgi> /blitz-support