Matlab incompatibility bug in ifft

[Ben Abbott]

On Wednesday, February 27, 2008, at 11:37AM, "Ryan Hinton" <iobass at email.com> wrote:
>Ben Abbott wrote:
>> On Monday, February 25, 2008, at 01:23PM, "Ryan Hinton" <iobass at email.com> wrote:
>>   
>>> Matlab supports an extra argument to the ifft command to force Hermitian 
>>> symmetry in the argument and so produce a real output.  This should make 
>>> the IFFT calculation faster and save an abs() call at the end.  (I am 
>>> implementing fast convolution of real signals, so I know the result must 
>>> be real also.)
>>>
>>> Octave example:
>>> ret = abs(ifft([1 + eps*1j;1 + eps*1j], [], 1));
>>>
>>> Matlab-supported shortcut:
>>> ret = ifft([1 + eps*1j;1 + eps*1j], [], 1, 'symmetric');
>>>
>>> This seems like a simple fix/enhancement, but I got lost in the code.  I 
>>> am happy to provide more information.  Or I can try a fix if someone can 
>>> help me figure out generally what needs to be done to which files.  I'm 
>>> familiar with m-code, but not with all the conventions and internal data 
>>> types of Octave's implementation.
>>>
>>> Thanks!
>>>
>>>     
>>
>> I don't have the c/c++ skills need to do this, but I checked the FFTW website
>>
>> http://www.fftw.org/
>>
>> and it indicates that this functionality is available. All that is needed is to modify the fft/ifft routines to respect the 'symmetric' option.
>>
>> Ben
>>   
>Yes, but how to do this?  I see FFT-like functions in five files,
>
>CMatrix.cc
>CNDArray.cc
>dMatrix.cc
>dNDArray.cc
>oct-fftw.cc
>
>but I can't figure out the first order of business: where are the 
>command line arguments?  In other words, if I add a 'symmetric' argument 
>at the end, where does this show up?
>
>---
>Ryan Hinton
>iobass at email.com
>

I'm unfamiliar with the code, but it looks like the right place to start is in fft.cc, fft2.cc, fftn.cc, etc.

Ben