C / C ++ non-GPL FFT library

I am looking for a licensed FFT library that allows me to incorporate it into commercial C ++ software. Free would be nice. (Intel IPPS is $ 199 per head, which is a bit expensive for the number of developers who will compile it.)

Does anyone have any good experience with such an FFT library?

I found this page, which lists many libraries: http://www.fftw.org/benchfft/ffts.html

FFTW has a GPL license, so it will not work. Any good alternative?

+47
c ++ c gpl signal-processing lgpl
Jan 20 '09 at 21:25
source share
9 answers

Kissfft

Some tests compared to fftw:

http://www.fftw.org/speed/Pentium4-2.4GHz-gcc/

EDIT: EigenFFT - C ++ interface for KissFFT as an unsupported part of Eigen.

+20
Jan 20 '09 at 22:13
source share
β€” -

Check out the FFTS. It has a BSD licensing license, and it is faster than FFTW, IPP, KissFFT, Apple vDSP, etc. On ARM and x86.

http://github.com/anthonix/ffts

+19
May 9 '13 at 22:43
source share

Laurent de Soras FFTReal is very fast, and LGPL'd means that you can use it in commercial software without having to pay a license and not make your software open source. You just need to assure the author if you use it in your software (and he deserves it!).

+4
Jan 20 '09 at 23:18
source share

Do you need sources? Do I need to be free, like in beer?

If not, the Intel IPP library at least gives you excellent performance, assuming your platform is among the supported ones.

+3
Jan 20 '09 at 21:34
source share

As Nick Reiman said, you can use FFTReal and respond to the Brone objection, the license has been changed to WTFPL, which is more flexible.

+3
Nov 04 '10 at
source share

According to the FFTW documentation, a non-GPL license is also available (presumably for $). You should contact them for more details.

+2
Jan 20 '09 at 21:57
source share

I found commercial FFTW not available. (They wanted a one-time fee of about $ 8,000, which I assume is for a commercial license.) You may have found that the Intel library is not as expensive as you think. We use Intel MKL (Math Kernel Library), and since it is intended for redistribution (with static or dynamic linking) as part of a commercial library, you can only get a license for a few developers who actively use It. To do this, all those developers should be placed in a separate library (since under license other developers cannot directly use the Intel library, including headers). The only caveat is that the open parts of this library should add significant functionality beyond the Intel library (i.e., creating an FFT wrapper and using it as a library that other users might receive would violate the license agreement). (Interpret the license in the sense that non-math developers are the β€œcustomers” of math developers.)

Just read the license agreement carefully and decide if you can apply this approach to your library of choice (IPPS).

+2
Jan 25 '09 at 17:40
source share

You can see the numerical recipes . It is not free, but cheaper IPP :). I used NR, but not FFT.

0
Jan 20 '09 at 10:50
source share

Read it again on Intel IPP, you only need one developer license @ $ 199 to install it on one machine. Any other developer can still compile it, they will not need to install the entire Intel IPP, only the corresponding headers and libraries.

0
Nov 04 '10 at 12:38
source share



All Articles