Gregory Lee

If you import from `numpy.fft.fftpack` instead of from `numpy.fft` you will get non-MKL versions. That is what we ended up doing in the [pyFFTW test suite](https://github.com/pyFFTW/pyFFTW/blob/5bf1dda837dcef722ee14b349834a701e4d53592/test/test_pyfftw_base.py#L42-L47). However, be aware that...

Yeah, skipping if ``np.fft.fft.__module__.startswith('mkl')`` sounds reasonable.

> Ok, fix for that coming along. Sadly it seems that the function you mentioned is not 100% valid for e.g. rfft so I need to special case that. This...

Related to this, the following work might eventually remove the need to maintain multiple FFT backends in ODL itself: We currently have a student working on support for multiple FFT...

> It should be mentioned that there is already the pyfftw numpy compatibility interfaces: https://hgomersall.github.io/pyFFTW/pyfftw/interfaces/interfaces.html the problem with them was that they do not provide optimal usage in terms of...

> Thanks Gokul! > cc @grlee77 (in case this is of interest wink) Yes, thank you @gokul. For context, I have been working on CUDA-based implementations of classical (i.e. not...

Hi @codesteller. It sounds like your use case is primarily for 1D signals? I am a maintainer at PyWavelets and had worked on a new package for discrete wavelet transforms...

Also, I don't think we currently use them anywhere in cuCIM, but this same implementation from #317 can also be used to accelerate CuPy's [percentile_filter](https://docs.cupy.dev/en/stable/reference/generated/cupyx.scipy.ndimage.percentile_filter.html) or [rank_filter](https://docs.cupy.dev/en/stable/reference/generated/cupyx.scipy.ndimage.rank_filter.html) just by changing...

no need to immediately review this for the 22.08 release since we don't ship the benchmark scripts

I recently made a notebook for stain normalization and can make a PR for it soon.