support mir.fft : thin wrapper around fftw for ND convolution
timotheecour opened this issue · 4 comments
The goal should be to be as efficient as possible (eg by providing a user friendly wrapper around fftw; note that https://dlang.org/library/std/numeric/fft.html is slow compared to fftw IIRC)
Duplicate of #45. Mir will have its own fft with the same spead as fftw (GLAS is fast as OpenBLAS, so I think we can do our fftw analog in D).
Note that the dplug:dsp package has a FFT that works with real inputs (public domain + boost license).
Might be used for 1d convolution only.
@ p0nce:
well there is already https://dlang.org/phobos/std_numeric.html#.fft but it's much slower than fftw.
@9il:
fftw is quite a large library, matching the speed would probably take a lot of effort/time; in the meantime having user friendly wrappers would be nice to have
fftw is quite a large library, matching the speed would probably take a lot of effort/time; in the meantime having user friendly wrappers would be nice to have
Agreed (but time constraints)