FFT performance depends a lot on the input size. Since we have the freedom to pad the ends, it's worth it to pad up to a size that gives good performance, which is when you have a simple prime factorization.
This is the speed of numpy.fft using padding with next_fast_fft, or using no padding.
FFT performance depends a lot on the input size. Since we have the freedom to pad the ends, it's worth it to pad up to a size that gives good performance, which is when you have a simple prime factorization.
This is the speed of numpy.fft using padding with![pyplots](https://cloud.githubusercontent.com/assets/641278/5545121/3ce93cc2-8ace-11e4-841d-540c00400b3b.png)
next_fast_fft
, or using no padding.(script: https://gist.github.com/rmcgibbo/7a45fe4a42cffd93bd02)