Bit-sliced storage for finite-field vectors - #251

Open

JoeyBF wants to merge 27 commits into

claude/finite-field-bit-storage-fyb15c

Commits on Jun 29, 2026

Phase 0: bit-sliced storage prototype + benchmark gate
claude
committed
f9e8926
Phase 0: record bit-slice benchmark gate results
claude
committed
763f63b
Phase 0b: tighten bit-slice generic kernel, refresh results
claude
committed
c792467
Phase 0b: dispatch bit-slice generic kernel to const-generic K
claude
committed
a91f374
Phase 0b: record const-K dispatch benchmark results
claude
committed
a37a1bd
Phase 1: introduce group-layout seam in FieldInternal (no-op)
claude
committed
b65ddf6
Phase 2: bit-sliced storage for all prime/extension fields
claude
committed
4aa4714
Phase 2: fast group-kernel path for bit-sliced slice addition
claude
committed
550725d
Phase 2: bit-sliced slice add handles equal nonzero lane offsets
claude
committed
a79e6cb
Phase 2: masked plane-circuit add for bit-sliced slice boundaries
claude
committed
ce635c3
Phase 2: cut per-call overhead in bit-sliced kernels
claude
committed
feff517
Phase 3: 6-gate F3 bit-slice add circuit
claude
committed
3ce4ab6
Phase 3: flat F5 bit-slice add/scale circuits
claude
committed
eb74407
Phase 2: fix bit-sliced shl_assign and augmented-matrix segment padding
claude
committed
f7624e9
Phase 5: remove dead odd-prime shift branch and orphaned num_limbs
claude
committed
81c4351
Phase 5: unify slice add, remove add_shift_* (general per-plane shifted add)
claude
committed
3399469
CI: rustfmt and fix a private intra-doc link
claude
committed
64b0aba
Phase 5: remove Phase 0 prototype scaffolding; clippy fixes
claude
committed
a9e0533
Fix CI lint/docs and harden bit-sliced slice ops
claude
committed
d88b76a
Run nightly rustfmt on bit-sliced group iteration
claude
committed
08c9ff6
Address review: bit-sliced add_truncate, empty range, ragged from_vec
claude
committed
c25446d
Rewrite p_masks loop with explicit conditional
claude
committed
8e81da3
Simplify pmask: Limb::from + wrapping_neg
claude
committed
4c67bbc

Commits on Jun 30, 2026

Credit Carl McTague for bit-slicing idea and F3 circuit

claude
and
mctague434
committed
d754fae
Fix two inaccurate comments in bit-sliced vector code
claude
committed
781b7a7

Commits on Jul 1, 2026

Trim redundant comments in bit-sliced code
claude
committed
12d6f02

Commits on Jul 16, 2026

Replace F5 add with Carl McTague's 17-gate circuit

claude
and
mctague434
committed
1d18af0