Repository navigation
Figure out which Kyber primitives ML-KEM can reuse #8
Description
ML-KEM uses the same NTT, Keccak, CBD and polynomial encoding as Kyber, so we shouldn't write them twice. But changing shared code risks changing Kyber output.
- Go through
fips202,ntt,reduce,cbd, poly encoding and compression, and list which ones ML-KEM can use as they are - Add a parameter type generic over k (const generic or a trait) for the new code. I don't want a fourth copy of the
_512/_768/_1024functions - Leave the legacy code alone apart from moving things around
Done when: the golden vectors from #6 still pass and the PR has a short write-up of what's shared and what isn't.
Depends on: #6. Plan: WP-2.
Part of #3.
Activity
Zikithezikit commentedon Oct 30, 2026
Done in #28, merged into dev-v2. The golden vectors still pass, and the write-up of what is shared and what is not is the module documentation at the top of src/mlkem/mod.rs (summary in the PR description).
Short version: Keccak, reduce, ntt, cbd2/cbd3, poly_tobytes, Compress/Decompress for d = 1, 4, 5, 10 (and Decompress for 11) and rej_uniform are shared as they are. Not shared: poly_frombytes (no mod q reduction), the legacy d = 11 compression (off by one at 1875 and 3125, logged on #18), poly_frommsg, verify/cmov and poly_getnoise_*. Added: ParamSet, PolyVec, sample_ntt, and an exact Compress_11 for ML-KEM-1024.
Open findings from this work are on #18. Next is #9.