Useful informations

Latency

TODO

Fast Fourier Transform

Intel Math Kernel Library (MKL)

The default fft backend is the Intel Math Kernel Library. To use this, you just need to download it from Intel’s website here: https://software.intel.com/en-us/mkl

FFTW

If for some reason you don’t want to use the MKL, we also provide a way to use the fftw for linux and OSX. All you need is to disable the use of the MKL in the CMake options:

cmake -Drtff_use_mkl=OFF ..

We also provide a way to leverage the fftw wisdom principle. See http://www.fftw.org/fftw3_doc/Wisdom-Import.html#Wisdom-Import for more information.

To enable this, add another option to the cmake command:

cmake -Drtff_use_mkl=OFF -Drtff_fftw_use_wisdom=ON ..

The wisdom consist of a file containing informations on how to compute a faster transform. Its generation takes a very long time (up to a couple of minutes) but it speeds up the fft computation once initialized.

In rtff, we look for files named rtff_[fft size].fftw in the current folder. If none can be found, the file will get generated so it can be loaded during the next run.

If the intial delay isn’t acceptable in your case, you can always generate the wisdom file yourself and name it correctly in you run directory. This way you can take advantage of the speed improvement without worring about the initialization duration.