zywina/cfftpack

C translation & wrapper of Fortran FFTPACK library with a few extras.

cfftpack

C translation & wrapper of Fortran FFTPACK library with a few extras.

Having wrapped components of FFTPACK several times, both personally and professionally I finally decided to do a proper wrapper for the full thing, and open source it with a permissive license, so I could use it for both personal and commercial projects. If other people find it useful that is cool too.

Current status is BETA. Most things work except for stride modifications on some calcs.

Those with a quantitative finance background might find the test directory interesting. It contains some work from as yet unpublished papers of mine.

cfftpack's Issues

TODO: add support for 2D convolution?

First, the library looks great as is. Nevertheless, adding a multiplication function that performs element-wise multiplication between two 2D arrays transformed with fft2() will extend the usability of the library to a greater audience, specially for those working on Image Processing applications on C/C++.

Are you still working on this?

Single-precision literals and using float and double together

All real literals in fftpack.c are double precision such as 2.0 or 1.0, except in 4 different instances we have .5f instead which is a bit odd.

Also in case anyone is interested, since I needed to be able to do both float and double FFTs in the same program I took fftpack.c, branched it into fftpack_float.c and fftpack_double.c, removed fft_real_t, removed the c__1 and c__2 globals (I put them inside the functions that need them instead to avoid having the same globals declared twice), removed everything related to the complex keyword and the other pointless macros, replaced fft_complex_t with fft_complexf_t / fft_complexd_t, then for the single precision version I converted the trigonometry function calls and the literals to single precision, and minor things like using the value of pi instead of calculating it from atan, made separate header guards, gave function names suffixes, all so that you can include both fftpack_float.c and fftpack_double.c in the same C file as I did here without conflict.

cfft2b_() described as "2D FFT"

In fftpack.h the comments describe cfft2b_() as "2D FFT" and cfft2f_() as "2D IFFT". Isn't it the opposite, since b stands for backward and f for forward? The FORTRAN documentation states that CFFT2B "is known as the backward transform or Fourier synthesis, transforming from spectral to physical space", making it the IFFT.

Either way, bang up job, I'll soon be using cfftpack in one of my programs to speed up convolution between image layers.