Resolve Compare FFT implementations #62

[einar-taiko]

This resolves https://github.com/taikoxyz/zkevm-circuits/issues/62

• Fast-forward main to main from PSE.

• Rebase Brecht's FFT fork on main

• Rebase Scroll's FFT fork on main

• Factor out redundancies between forks

• Reuse tests from branches

• Verify cargo test

• Please cargo check --all-targets

• Please cargo clippy --all-targets

[Brechtpd]

Diff much improved!

[einar-taiko]

@Brechtpd I have reverted as much as possible to produce a minimal diff -- including the things you mentioned.

I have also chosen recursive_fft as the default now.

Please have another look at the diff and see if there are further changes you would like.

[Brechtpd]

Ah nice diff now! Just some comments, overall looks great now!

[einar-taiko]

@Brechtpd Please confirm which implementation you would like as default, keeping in mind that recursive is slightly faster for k=25 while it runs OOM for k=29. Besides that, I'd say the PR is ready.

[Brechtpd]

Just some nitpicks mostly about the diff, otherwise LGTM!

[einar-taiko]

All suggestions implemented.

@Brechtpd Please confirm which implementation you would like as default, keeping in mind that recursive is slightly faster for k=25 while it runs OOM for k=29. [...]

@Brechtpd Please let me know which you prefer as default before we merge. It currently defaults to baseline.

[Brechtpd]

@Brechtpd Please let me know which you prefer as default before we merge. It currently defaults to baseline.

Let's go with the parallel one, I think that one will is safer when running with a non power of 2 number of cores/hyper threading.

[einar-taiko]

Let's go with the parallel one.

Done.

[Brechtpd]

Still two diff changes left that I unresolved, see above. After that were done!

[einar-taiko]

Still two diff changes left that I unresolved, see above. After that were done!

@Brechtpd Not sure what went wrong last night. The last two should now be reverted in 5bc5ca5.