Poseidon 2

[mpenciak]

Poseidon 2 chip for Loam.

Generic in width using hybrid-array

The structure of the API around the chip is in a good spot I think.
The tests don't currently pass, I'm debugging!

Please take a look to make sure everything is as one would expect from the precompile

Note 1:

In order for this to work I had to fork Plonky3 and make a couple small changes

1. (uncontroversial) Add a few extra supported arities.
2. (potentially controversial) Change the visibility of the Permutation trait so I could implement a new version to support fixed round constants, and not those generated by the rng in Plonky3.

Note 2:

I added the AlignedBorrow in loam-macros pretty early on in the process. I don't think it's strictly necessary anymore but it could help in other spots...

Note 3:

I'm not sure what to do about the sage script... It doesn't make much sense to include it in the repository, but I also don't know if it's good practice for people to take us on our word that we generated the constants correctly.

[adr1anh]

• Made some suggestions to make it more "Air-like" and avoid allocations in the eval function at ah/poseidon-air.
• I think you were missing the initial linear layer.
• We should find a way to return the constants without allocating a vector.
• The config could have default methods for returning the ranges of the different rounds. fn external_first_range() -> Range<usize> etc.

[adr1anh]

Comments about Air

• It's common for the eval function to define all the flags and bools at the start, so it's clear when later constraints are being used.
• The is_internal and is_external can be derived from the sum of the rounds flags.
• The application of the constants to the rounds cannot have an if. Instead, you need to conditionally select the constants using local.rounds, and use another selector to conditionally add the constants (is_external, is_init, is_internal)

[arthurpaulino]

Great work 🎉
Left some higher level comments related to maintenance

[mpenciak]

Updates: I applied the suggests in the comments, and fixed a small error in the internal matrix action to fix the tests.

The only couple outstanding issues to address before I think this is ready to approve:

1. What to do with the sage script? I don't think it really fits in this repo, but we could add a public lurk-lab gist with the script so people can verify/audit it and link to it in the constants.rs file.

2. What to do about the mp/poseidon_fix branch on lurk-lab/Plonky3? I could create a PR and merge into our Plonky3 repo, but I don't know if that adds a lot of friction to keeping our version of Plonky3 in sync with the upstream one.

[huitseeker]

Swooping in to block on the personal fork branch.

[wwared]

This is great! I have a tiny suggestion below (feel free to ignore). Regarding the other two points mentioned, here's my two cents:

1. I think the .sage script would not necessarily be out of place in this repo in some additional folder in the repo (assets/ or scripts/ or even docs/) and would vote for that, but I think the public-gist solution is also perfectly acceptable.
2. We definitely want the changes of mp/poseidon_fix to at least be merged into our fork, they are very minor and are unlikely to be a maintenance burden. I think upstream would be open to them as well, but we can do that later.

[huitseeker]

Please use workspace dependencies. To do this, you can use the cargo autoinherit tool.

[adr1anh]

Leaving some comments to be converted to issues

[adr1anh]

![derive(Default)]

[mpenciak]

Because C: PoseidonConfig doesn't have a blanket Default implementation (and it shouldn't, because the round constants need to be provided by hand) this doesn't work and I had to do it by hand.

[huitseeker]

Good for me!

[wwared]

Looks good! Needs a final rebase/conflict resolve before merge

[mpenciak]

Rebased!