We have optimized a lattice-based signature algorithm: ml-dsa
Changes:
-
Use an optimized SHAKE128 implementation
-
Optimize the matrix-filling strategy
The changes are small and easy to review. Benchmarks show approximately 60–70% of the original cycle count.
We have also provided CKB-VM optimized implementations of SHAKE128, SHAKE256, SHA3-256, and SHA3-512: