Update to limitations

[smu160]

@astral4

Hi,

Thank you for creating and maintaining this benchmark repo. Yesterday, we released v0.2.0 of phastft. This release includes support for single precision signals.

Similar to fftw, we now have fft_64 and fft_32. More information can be found in the docs. Despite the changes to the API,, we decided to take a more conservative approach and not do a major SemVer bump. Personally, I am hoping to only reach v1.0.0 once the project is a bit more mature and stable.

As always, please feel free to reach out with any questions, comments, or concerns.

I will also add a changelog very soon to acknowledge contributions, suggestions, etc.

Thank you!

Best,
Saveliy

[astral4]

Thank you for letting me know! I will update the benchmarks as soon as I can.

[smu160]

Just an FYI: there are two performance related issues that I expect to come up when you run the benchmarks using v0.2.1.

1. The inverse FFT will, potentially, be significantly slower this time around. The reason being is here. This solution is temporary -- I just need to figure out why the original implementation was causing output issues.

2. Since you run benchmarks on a machine with a ARM cpu, the phastft will, potentially, have performance regressions. The cause for this is that we moved away from requiring building from source. We now use a dynamic dispatch solution using multiversion thanks to a contribution from the multiversion author, @calebzulawski. You can see the usage here. We should be able to resolve this issue by simply adding more target architectures/features.

Thank you for all your hard work!

Best,
Saveliy