Forums

FFTW 3.1.2 - fftw_plan_many_dft

Started by solonas13 May 18, 2008
Hello guys,

I wanted to ask something concerning FFTW 3.1.2
I want to compute a 2D Fourier Transform of a 64x64 array. For research
reasons I need to calculate firstly 64 transforms, thus the rows, and then
in order to transpose between the two dimensions I NEED to use the strided
calls to check their performance. This is what I do:

FOR the x dimension
-------------------

fftw_plan_many_dft(rank, n, howmany, input, inembed=NULL, istride, idist,

	output, onembed, ostride=NULL, odist, FFTW_FORWARD, FFTW_MEASURE);

where rank=1, n[0]=64, howmany=64, istride=ostride=1, idist=odist=64

and to compute the 2nd dimension I pass as an input the output from above

FOR the y dimension
-------------------

fftw_plan_many_dft(rank, n, howmany, output, inembed=NULL, istride2,
idist2, final, onembed=NULL, ostride2, odist2, FFTW_FORWARD,
FFTW_MEASURE);

where rank=1, n[0]=64, howmany=64, istride2=ostride2=64, idist2=odist2=1

Any ideas why am I getting this all wrong? From my tests I thing that the
wrong arguments are those concerning the stride and dist but they seem
correct... Please help

Thank you in advance


I think the better and faster approach will be to use fft2 function
available in octave. Octave is a free equivalent of matlab and can
be installed on windows and linux machines.

y = fft2(a, m, n);

Regards
Bharat Pathak

Arithos Designs
www.Arithos.com

DSP Design Consultancy and Training Company.
"bharat pathak" <bharat@arithos.com> wrote in message 
news:qY-dnW3XBLSVea3VnZ2dnUVZ_hOdnZ2d@giganews.com...
>I think the better and faster approach will be to use fft2 >function > available in octave. Octave is a free equivalent of matlab > and can
> be installed on windows and linux machines.
^^^^^^^^^^^^^^^^^^^^ Only if you first install Cygwin. For this reason, Octave is a big disappointment and is in no way a replacement for Matlab (or FFTW).
On May 18, 9:41 pm, "solonas13" <solona...@hotmail.com> wrote:
> Hello guys, > > I wanted to ask something concerning FFTW 3.1.2 > I want to compute a 2D Fourier Transform of a 64x64 array. For research > reasons I need to calculate firstly 64 transforms, thus the rows, and then > in order to transpose between the two dimensions I NEED to use the strided > calls to check their performance. This is what I do: > > FOR the x dimension > ------------------- > > fftw_plan_many_dft(rank, n, howmany, input, inembed=NULL, istride, idist, > > output, onembed, ostride=NULL, odist, FFTW_FORWARD, FFTW_MEASURE); > > where rank=1, n[0]=64, howmany=64, istride=ostride=1, idist=odist=64 > > and to compute the 2nd dimension I pass as an input the output from above > > FOR the y dimension > ------------------- > > fftw_plan_many_dft(rank, n, howmany, output, inembed=NULL, istride2, > idist2, final, onembed=NULL, ostride2, odist2, FFTW_FORWARD, > FFTW_MEASURE); > > where rank=1, n[0]=64, howmany=64, istride2=ostride2=64, idist2=odist2=1 > > Any ideas why am I getting this all wrong? From my tests I thing that the > wrong arguments are those concerning the stride and dist but they seem > correct... Please help > > Thank you in advance
The question that came to my mind is: why not just use FFTW's built-in 2-D transform functionality? It's easier to use than the advanced interface you're working with. At one point I did understand the meaning of all of the arguments (I think the documentation is a bit terse on that topic), but my memory fails me now, so I'm not sure what you might have wrong. fftw_plan fftw_plan_dft_2d(int nx, int ny, fftw_complex *in, fftw_complex *out, int sign, unsigned flags); Jason
>Only if you first install Cygwin. For this reason, Octave is a >big disappointment and is in no way a replacement for Matlab >(or FFTW).
They have octave-forge version which does not require cygwin. I have it running on XP machine. Cygwin not needed.