[SciPy-user] Why are there two FFTs?

Travis Oliphant oliphant.travis at ieee.org
Fri Jan 6 15:11:38 CST 2006


Sebastian Haase wrote:

>On Friday 06 January 2006 11:09, Travis Oliphant wrote:
>  
>
>>Gary wrote:
>>    
>>
>>>scipy.fftpack.*   and numpy.dft.*
>>>
>>>are they different under the hood
>>>      
>>>
>>Yes.  scipy.fftpack can link to fftw and/or djbfft for example which can
>>be faster (but also have more install difficulties...).  Also the
>>extension module is f2py generated.
>>
>>numpy.dft always uses an f2c'd version of fftpack and a hand-written
>>extension module.
>>
>>-teo
>>    
>>
>
>Is there any support for "single precision" (float32,complex32) fft ?
>We consider this quite important since our 3D image can already be on the 
>order of 500MB ...
>
>  
>
No, I don't think this is supported yet.  But, it could be added for 
scipy.fftpack.   I don't think we will add it to numpy.

Otherwise, you will need to call out to the library yourself (use 
numpy.f2py to help...)

-Travis



More information about the SciPy-user mailing list