search

IntelPython / mkl_fft

NumPy-based Python interface to Intel (R) MKL FFT functionality
BSD 3-Clause "New" or "Revised" License
65 stars 16 forks source link

Provide Dask interface to mkl_fft #5

Open jakirkham opened 6 years ago

jakirkham commented 6 years ago

Would be nice if a simple dask interface were provided to mkl_fft (much like the numpy one currently). In fact it could be built right on top of the _numpy_fft module by wrapping these functions using fft_wrap just as it is done in Dask except using the one's in _numpy_fft instead.

jakirkham commented 6 years ago

Any thoughts on this, @oleksandr-pavlyk?

oleksandr-pavlyk commented 6 years ago

@jakirkham I think dask interface is a great idea. I am not clear as to what exactly should be done in _numpy_fft.

Are you proposing to add _dask_fft.py to the package, which would export these fft_wrapped function if dask is found in the environment?

jakirkham commented 6 years ago

Thanks for the follow up. Glad to hear this works.

I am not clear as to what exactly should be done in _numpy_fft.

Don’t think anything has to be done to _numpy_fft. That just works well for this purpose. Namely fft_wrap works well with FFT functions adhering to the NumPy API.

Are you proposing to add _dask_fft.py to the package, which would export these fft_wrapped function if dask is found in the environment?

Yep, exactly.

Any thoughts on how we name things in the public API? Also any thoughts on what would be adequate testing?