[flyq]

Summary

Ports PR #4 of suryuhh/salt-optimizations (commit e7e1e43): the committer's fixed-base multiplications, eight per vector register, over smaller tables.

• Committer::mul_index_batch is new. On a host with AVX-512 IFMA it runs eight multiplications per register; elsewhere it maps mul_index.
• The trie update issues its multiplications per chunk through it.
• On a host with IFMA the shared committer is built at window 7 instead of 11. The external commit gives 39.4 MB of table data for 256 bases at window 7, against 403 MB at window 11.

Hosts without IFMA keep window 11 and the scalar path.

Stacked on the compact storage PR (liquan/perf/compact-witness-storage). Last of the four.

Changes on top of the external commit

• msm and msm_dense run in lanes too. At window 7 one mul_index is 37 additions instead of 24. The external change moved the trie update to the batch, but Committer::msm and msm_dense still multiplied one scalar at a time. Proof creation (the IPA rounds, the dense polynomial commitment) and the node-polynomial cache's admission check go through those two, so each would have slowed by about half on IFMA hosts. Each thread's terms now go through mul_index_batch there, in tasks of whole lane groups. Hosts without lanes keep the scalar loop as it was.
• The IPA's blinding term joins the round's msm. It was a lone mul_index per side per round.
• The shared committer keeps the G ++ [Q] tables of perf: faster witness proofs and proof checks #152. The only cherry-pick conflict: DEFAULT_CRS.committer(platform::precomp_window_size()).
• The table point layout is asserted. mul_index8 gathers x at +0 and y at +32 of an EdwardsAffine; a different layout now fails to compile.
• Window 7 is in the default equivalence test. It was only checked by an ignored bench.
• Four line-pinned suppressions in trie.rs are re-pinned (905/911/914/916 to 912/918/921/923). The code at those sites did not change.

Testing

Passing on this head, on an Apple Silicon machine:

• cargo fmt --all -- --check; cargo clippy --all-targets -- -D warnings, native and --target x86_64-unknown-linux-gnu
• cargo test; with test-bucket-resize (CI's settings); the random stress test; --no-default-features with and without test-bucket-resize
• cargo check -p salt --target riscv64imac-unknown-none-elf --no-default-features
• cargo check --all-targets --target i686-unknown-linux-gnu
• cargo-mutants on the diff: 8 of 8 viable mutants killed, 2 unviable. Spec-gate: 1 of 1 killed. Suppression hygiene passes.
• stateless-validator main (c46a541): cargo check --workspace --all-targets with its salt dependency patched to this head.

New tests run on every host: mul_index_batch against mul_index at windows 11 and 7, the batched sum against the scalar sum and the reference MSM, and the task size. create_with_precomp_matches_create covers the blinding term.

Not run:

• Any AVX-512 path, including the lane msm above. No such CPU or emulator on the machine used; compiled and linted for x86_64 only. On a host without lanes the new tests only exercise the scalar fallback of mul_index_batch. The Test Suite (AVX-512 IFMA) job is their first real execution.
• Benchmarks. The claim that proof creation does not regress at window 7 is by construction, not measured. perf-regression runs update 10000 KVs on a runner with IFMA, so the trie update number arrives with this PR. salt/benches/proof.rs (witness creation) is not covered by it and needs a run on a host with IFMA against main.
• mega-reth against this head.
• The i686 test job.

Merging

Please keep this trailer in the squash message:

Co-authored-by: suryuhh <suryuhh@users.noreply.github.com>

🤖 Generated with Claude Code

[github-actions]

Mutation testing - PASS

Mutation score: 100.0% (1/1 viable mutants killed)

• caught: 1
• survived, real gaps: 0
• timed out, inconclusive: 0
• suppressed: 0
• unviable: 0
• timeout total: 0

No unsuppressed survivors or timeouts remain.

[github-actions]

Performance Benchmark Comparison

Compared 5 benchmark(s) against the latest main baseline.

Detailed Comparison

Benchmark	Baseline Throughput (Kelem/s)	New Throughput (Kelem/s)	Change
update 10000 KVs/1 threads	83.52	174.16	+108.53%
update 10000 KVs/2 threads	160.32	323.25	+101.63%
update 10000 KVs/4 threads	299.94	539.85	+79.99%
update 10000 KVs/8 threads	490.69	804.66	+63.99%
update 10000 KVs/16 threads	617.42	916.97	+48.52%

[github-actions]

Mutation testing - PASS

Mutation score: 100.0% (6/6 viable mutants killed)

• caught: 6
• survived, real gaps: 0
• timed out, inconclusive: 0
• suppressed: 0
• unviable: 4
• timeout total: 0

No unsuppressed survivors or timeouts remain.

[codecov]

Codecov Report

❌ Patch coverage is 82.64984% with 55 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
banderwagon/src/ifma_committer.rs	77.41%	34 Missing and 1 partial ⚠️
banderwagon/src/salt_committer.rs	85.34%	13 Missing and 4 partials ⚠️
banderwagon/src/platform.rs	50.00%	2 Missing and 1 partial ⚠️

Files with missing lines	Coverage Δ
banderwagon/src/lib.rs	70.00% <ø> (ø)
ipa-multipoint/src/ipa.rs	99.39% <100.00%> (+<0.01%)	⬆️
salt/src/trie/trie.rs	98.22% <100.00%> (+0.01%)	⬆️
banderwagon/src/platform.rs	50.00% <50.00%> (ø)
banderwagon/src/salt_committer.rs	93.07% <85.34%> (-4.45%)	⬇️
banderwagon/src/ifma_committer.rs	77.41% <77.41%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.