Repository navigation

New 3-way batched Keccak-f1600 (x3) improves ML-DSA performance by 2.5 to 13.0% on Neoverse V3 - #1489

Draft

soumya2094 wants to merge 8 commits into

mainfrom

x3-hybrid-keccak-f1600

Draft

New 3-way batched Keccak-f1600 (x3) improves ML-DSA performance by 2.5 to 13.0% on Neoverse V3#1489
soumya2094 wants to merge 8 commits into
mainfrom
x3-hybrid-keccak-f1600

Commits on Oct 25, 2026
- FIPS202/AArch64: Add symbolic Keccak sources and SLOTHY Makefile
  soumya2094
  committed
  772e436
- FIPS202: Add 3-way batched Keccak-f1600 with an AArch64 hybrid backend
  soumya2094
  committed
  a7b795b
- HOL-Light: Prove the x3 hybrid Keccak-f1600 kernel
  soumya2094
  committed
  9905ad3
- Integration of the 3-way batched kernel with MLD_CONFIG_USE_X3
  soumya2094
  committed
  8d76177
- test: Add an x3 test configuration and update component benchmarks
  soumya2094
  committed
  8dc139b
- CBMC: Add proofs for the 3-way batched Keccak and samplers
  soumya2094
  committed
  f71add1
- CI: Test the x3 configuration
  soumya2094
  committed
  c2d8b2a
- bench: Benchmark the x3 Keccak on Graviton 5
  soumya2094
  committed
  b622325