fftw interface with complex.h in c

3
votes

I am using this code to create a 2D array having data type complex (from complex.h in c). Then I want to find the fft of that array 'in place'. However it is giving segmentation fault, which I fear is due to improper casting of the pointer. How do we use complex data type for a fftw?

//a is a 2d array of size blockSize*NO_INPUTS

complex float* a=(complex float*)malloc(sizeof(complex float)*NO_INPUTS*blockSize);

//****** Put data in a*******//
//blah//
//blah//

//a has data now//

fftw_plan p;
p=fftw_plan_dft_2d(blockSize,NO_INPUTS,(fftw_complex*)a,(fftw_complex*)a,FFTW_FORWARD,FFTW_ESTIMATE);

fftw_execute(p);

fftw_destroy_plan(p);

Please note that I do not want to use fftw_complex or fftw_malloc. Thank you.

2
cfftwcomplex-numbers
It's been a while since I've used FFTW, but I seem to recall that fftw_complex defaults to complex if complex is defined. Regarding fftw_malloc, why don't you want to use it? These algorithms are highly dependent on the memory being byte-aligned. - Jim Clay
In addition to good points from @Jim above, I think there are some additional constraints on the size of your 2D array for in-place 2D FFTs - you need some additional storage per row I believe (one extra elements ?) - there's a section in the FFTW manual on this. - Paul R

2 Answers

3
votes

fftw_complex is equivalent to complex double, you are using complex float. That's probably the source of segfault.

Try including complex.h before fftw3.h you won't need to do any casts.

http://www.fftw.org/fftw3_doc/Complex-numbers.html

1
votes

FFTW defaults to using double precision floats by default. The simplest fix is just to change all instances of float to double. If, however, you want to continue using complex float for memory or other reasons then you should use single precision fftw as outlined here. You should change lowercase fftw in all fftw function and type names to fftwf. For example fftw_plan becomes fftwf_plan. On compilation, you should link to fftw3f instead of fftw3.