Update Kyber and Dilithium ARM code.

[xvzcf]

No description provided.

[xvzcf]

As previously noted, our implementations are from PQClean, which in turn takes it from the neon-ntt project.

[baentsch]

As previously noted, our implementations are from PQClean, which in turn takes it from the neon-ntt project.

Interesting to note -- what's the implication of this for us? Did PQClean not integrate the latest code from neon-ntt yet? Is neon-ntt itself not up-to-date compared to the ref implementation?

[dstebila]

PQClean doesn't have the latest code from neon-ntt.

[baentsch]

PQClean doesn't have the latest code from neon-ntt.

Is this still the case? Or would/should a fresh copy_from_upstream resolving #1506 also resolve this issue?

[mczraf]

PQClean doesn't have the latest code from neon-ntt.

Is this still the case? Or would/should a fresh copy_from_upstream resolving #1506 also resolve this issue?

I believe it is still the case. I've checked PQClean repo, and they are still using the original Neon implementation from Sep/2021.

It might be worth adding that upgrading this implementation to a newer version may lead to some additional benefits. For example, LibOQS Kyber's implementation is not marked as passing all Valgrind branch-on-secrets tests (see here). There's a chance this could be automatically fixed by simply resorting to the newer Neon implementation. PS: At the moment, I can't dedicate cycles to this effort myself unfortunately but I thought it would be helpful to comment here to reinforce that this seems like an interesting move for LibOQS.

Thank you all for this amazing project :)

Rafael Misoczki (speaking for myself)

[baentsch]

Thanks for the feedback, @mczraf ! PQClean/PQClean#493 seems to confirm --and be work to resolve-- this. Add: Ouch -- that's Falcon, not Kyber.

[SWilson4]

Fixed by #1512.