search

StingraySoftware / stingray

Anything can happen in the next half hour (including spectral timing made easy)!
https://stingray.science/stingray
MIT License
174 stars 144 forks source link

Do performance tests with pytorch.fft #547

Open dhuppenkothen opened 3 years ago

dhuppenkothen commented 3 years ago

Should try and do some performance tests with pytorch. I'm wondering if there are other recent insights from computer science that could make our code faster/better? Should do a bit of a literature search on what made the PyTorch folks implement FFTs.

For reference, see the PyTorch documentation

soma2000-lang commented 1 year ago

@abigailStev @matteobachetti @dhuppenkothen I think the torch.fft module was implemented mainly because

PyTorch FFT implementation provides support for parallelized FFTs using the torch.nn.DataParallel wrapper, which allows users to perform FFTs on multiple GPUs. This can be useful for accelerating FFTs on large datasets or for performing FFTs in real-time applications.

soma2000-lang commented 1 year ago

But I considering the fact how can we make stingray fast by using the torch.fft module.I would like to discuss more on this topic