feat(sumcheck): complete y-first cutover and recover perf by quangvdao · Pull Request #46 · LayerZero-Labs/hachi

quangvdao · 2026-04-07T23:24:41Z

Motivation

Technique 2's y-first cutover is now complete across both Stage 1 and Stage 2. We need to preserve the semantic switch to y-first ordering, keep the compact prefix fast paths, and recover the profiling regressions that showed up during the migration.

Summary

complete the Stage 1 plus Stage 2 y-first cutover, with no return to x-first compatibility shims
keep the compact witness and recursive opening-point flows aligned with y-first challenge ordering
recover the observed Stage 1 onehot regression by fusing the sparse y-stage full-table fold with next-round polynomial generation

Key changes

reorder Stage 1 inputs to y-first at the call sites and make the recursive opening point carry y-first challenges directly
store compact witness evaluations as x-outer, y-inner slices and thread that layout through Stage 1, Stage 2, the tree backend, and two-round prefix reconstruction
keep verifier-side Stage 2 m(x) sourcing explicit via Stage2MEvalSource and y-first ring-switch outputs
add a fused Stage 1 sparse-x/y handoff so the y-first cutover keeps onehot performance close to or better than main without reverting semantics

Performance

Benchmarks on the local profiling machine, using warmed reruns against a clean main worktree:

onehot nv32: commit 0.290s -> 0.190s, prove 0.833s -> 0.748s, verify 0.105s -> 0.095s
dense nv26: commit 8.640s -> 5.672s, prove 2.258s -> 1.830s, verify 0.0420s -> 0.0341s

Test coverage

cargo fmt -q
cargo clippy --all --message-format=short -q -- -D warnings
cargo test

Files/components impacted

src/algebra/shifted_eq.rs
src/protocol/commitment_scheme.rs
src/protocol/hachi_poly_ops/onehot.rs
src/protocol/ring_switch.rs
src/protocol/sumcheck/hachi_stage1.rs
src/protocol/sumcheck/hachi_stage1_tree.rs
src/protocol/sumcheck/hachi_stage2.rs
src/protocol/sumcheck/two_round_prefix.rs

Posted by Cursor assistant (model: GPT-5.4) on behalf of the user (Quang Dao) with approval.

Made with Cursor

Switch Stage 2 to the y-first witness layout and compute ring-switch m(x) on demand so the verifier no longer materializes m_evals_x. Keep Stage 1 x-first behind compatibility shims, including compact witness transposes, two-round-prefix updates, and wiring-layer challenge reordering, so recursive proofs continue to chain correctly during the Phase 1 transition. Made-with: Cursor

cursor · 2026-04-07T23:24:46Z

PR Summary

High Risk
High risk because it changes core cryptographic proving/verifying logic (sumcheck challenge ordering, witness table layout, and recursive opening-point derivation), where subtle ordering bugs could invalidate proofs or weaken soundness.

Overview
Completes the y-first challenge-order cutover across ring switch, Stage 1, Stage 2, and recursive level chaining, removing the prior x/y reordering shim and instead threading y-first challenges directly as the next-level opening point.

The compact witness layout is switched to x-outer/y-inner throughout, updating Stage 1/2 folding, prefix fast-paths, and the Stage1 tree backend accordingly; a new Stage 1 sparse x/y fusion path is added to regain performance after the ordering change. Verifier-side Stage 2 now takes an explicit Stage2MEvalSource for m(x) evals, ring-switch verifier input validation is tightened, and multiple tests are updated/added to lock in the new opening geometry and fused transitions.

^{Reviewed by Cursor Bugbot for commit 7350955. Bugbot is set up for automated code reviews on this repo. Configure here.}

github-actions · 2026-04-07T23:27:24Z

Benchmark Report

Latest run: 91fb27c
Message: Merge 7350955 into 9c8bbaf
Ref: quang/shifted-eq-dp
Main baseline: 9c8bbaf from merge-base on main.
Previous run: 02c8b9d from the previous successful PR update.
Binary: target/release/examples/profile.
Memory: maximum resident set size from /usr/bin/time on the benchmark process.

Single Polynomial x 32 Variables

Benchmark: 1-of-256 one-hot with 32 variables
Sparsity: each polynomial is 1-of-256 one-hot (equivalently, 1-sparse over 256 slots, density 0.39%).
Command: target/release/examples/profile with HACHI_MODE=onehot HACHI_NUM_VARS=32 HACHI_NUM_POLYS=1 HACHI_PROFILE_TRACE=0 HACHI_PROFILE_SPAN_CLOSES=0 HACHI_PROFILE_LOG=info HACHI_PROFILE_ANSI=0.

Metric	Main baseline	Previous run	Latest run	Unit
Setup	3.716	4.016	3.774	s
Commit	1.759	1.826	1.765	s
Prove (Hachi)	4.077	4.624	4.529	s
Prove (Total)	4.077	4.624	4.529	s
Verify (Hachi)	0.177	0.198	0.181	s
Verify (Total)	0.177	0.198	0.181	s
Max RSS	2834.2	2834.5	2834.9	MiB

Proof size: 73,712 B
Hachi fold bytes: 34,288 B
Tail bytes: 39,424 B
Proof framing bytes: 0 B
Hachi levels: 7
Tail shape: 78,848 elems at 4 bits/elem
Observed terminal state: w_len=78,848 with log_basis=4

Per-level parameters

L	Config	D	nA	nB	nD	lb	l1	m	r	δcommit	δopen	δfold	next w (ring)	next w (field)	planned bytes
L0	`D32-na3`	32	3	2	2	2	256	16	11	1	64	11	1,245,760	39,864,320	4,672 B
L1	`D32-na2`	32	2	2	2	2	256	13	8	1	64	10	98,334	3,146,688	4,352 B
L2	`D32-na2`	32	2	2	2	3	256	11	6	1	43	6	17,822	570,304	4,832 B
L3	`D32-na2`	32	2	2	2	4	256	10	5	1	32	5	6,113	195,616	5,216 B
L4	`D32-na2`	32	2	2	2	4	256	9	4	1	32	5	3,707	118,624	5,072 B
L5	`D32-na2`	32	2	2	2	4	256	9	3	1	32	4	2,880	92,160	5,072 B
L6	`D32-na2`	32	2	2	2	4	256	9	3	1	32	4	2,464	78,848	5,072 B

Per-level proof-size breakdown

L	total	y_ring	v	stage1 sc	interstage	s_claim	stage2 sc	next_w_commit	next_w_eval
L0	4,672 B	512	1,024	832	0	16	1,248	1,024	16
L1	4,352 B	512	1,024	704	0	16	1,056	1,024	16
L2	4,832 B	512	1,024	1,280	0	16	960	1,024	16
L3	5,216 B	512	1,024	1,728	32	16	864	1,024	16
L4	5,072 B	512	1,024	1,632	32	16	816	1,024	16
L5	5,072 B	512	1,024	1,632	32	16	816	1,024	16
L6	5,072 B	512	1,024	1,632	32	16	816	1,024	16

4 Polynomials x 30 Variables

Benchmark: 4 same-point 1-of-256 one-hot polynomials with 30 variables each
Batch: same-point opening of 4 polynomials, each with 30 variables.
Sparsity: each polynomial is 1-of-256 one-hot (equivalently, 1-sparse over 256 slots, density 0.39%).
Command: target/release/examples/profile with HACHI_MODE=onehot HACHI_NUM_VARS=30 HACHI_NUM_POLYS=4 HACHI_PROFILE_TRACE=0 HACHI_PROFILE_SPAN_CLOSES=0 HACHI_PROFILE_LOG=info HACHI_PROFILE_ANSI=0.

Metric	Main baseline	Previous run	Latest run	Unit
Setup	5.141	5.376	5.168	s
Commit	2.247	2.158	2.227	s
Prove (Hachi)	3.913	4.682	4.375	s
Prove (Total)	3.914	4.682	4.375	s
Verify (Hachi)	0.178	0.199	0.179	s
Verify (Total)	0.178	0.199	0.179	s
Max RSS	2892.0	2892.0	2892.5	MiB

Proof size: 75,248 B
Hachi fold bytes: 35,824 B
Tail bytes: 39,424 B
Proof framing bytes: 0 B
Hachi levels: 7
Tail shape: 78,848 elems at 4 bits/elem
Observed terminal state: w_len=78,848 with log_basis=4

Per-level proof-size breakdown

L	total	y_ring	v	stage1 sc	interstage	s_claim	stage2 sc	next_w_commit	next_w_eval
L1	4,352 B	512	1,024	704	0	16	1,056	1,024	16
L2	4,832 B	512	1,024	1,280	0	16	960	1,024	16
L3	5,216 B	512	1,024	1,728	32	16	864	1,024	16
L4	5,072 B	512	1,024	1,632	32	16	816	1,024	16
L5	5,072 B	512	1,024	1,632	32	16	816	1,024	16
L6	5,072 B	512	1,024	1,632	32	16	816	1,024	16

Posted by Cursor assistant (model: GPT-5.4) on behalf of the user (Quang Dao) with approval.

src/protocol/sumcheck/hachi_stage2.rs

Restore the fused stage-2 round-2 handoff so y-first prefix proofs stop rescanning the compact witness after the two-round-prefix transition, and clear the prefix state once that handoff completes to avoid stale-path reentry. Also narrow the temporary dead-code allowances introduced during the phase-1 y-first cutover by routing the verifier through the shared shifted-eq dispatcher and dropping now-unused test helpers. Made-with: Cursor

src/protocol/commitment_scheme.rs

Make stage 1 bind y-first and move the only coordinate reorder to stage-1 input so stage 2 consumes r_stage1 directly. Preserve the compact two-round prefix path and sparse-x handling while removing the old stage1-to-stage2 compatibility bridge. Made-with: Cursor

Remove the dead planner option builder so PlannerOptions only exposes configuration toggles that are still wired into the search flow. Made-with: Cursor

src/protocol/ring_switch.rs

Remove the unused shifted-eq evaluation helpers and the stale test that only exercised them so CI can keep treating dead code as an error. Made-with: Cursor

After the y-first cutover, recursive stage transitions can carry the sumcheck challenges directly as the next opening point. Remove the identity helper and the dead width bookkeeping it forced the prover and verifier to carry. Made-with: Cursor

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Reviewed by Cursor Bugbot for commit 54be014. Configure here.}

src/protocol/sumcheck/hachi_stage1.rs

Fuse the sparse y-stage full-table fold with next-round polynomial generation so the y-first Stage 1 cutover recovers the onehot regression without reverting semantics. Made-with: Cursor

Now that y-first is the only ordering, remove the _y_first suffix from reorder_stage1_coords, build_compact_s_table, and all related variable names. Deduplicate pad_compact_witness, advance_stage1_claim, and reorder helpers that were copy-pasted across three test modules. Delete the unused shifted_eq module. Made-with: Cursor

Made-with: Cursor

cursor bot reviewed Apr 7, 2026

View reviewed changes

src/protocol/sumcheck/hachi_stage2.rs Outdated Show resolved Hide resolved

cursor bot reviewed Apr 8, 2026

View reviewed changes

src/protocol/commitment_scheme.rs Outdated Show resolved Hide resolved

quangvdao added 2 commits April 10, 2026 13:45

refactor(planner): drop unused opt_sumcheck setter

76de9d4

Remove the dead planner option builder so PlannerOptions only exposes configuration toggles that are still wired into the search flow. Made-with: Cursor

cursor bot reviewed Apr 10, 2026

View reviewed changes

src/protocol/ring_switch.rs Outdated Show resolved Hide resolved

quangvdao added 2 commits April 10, 2026 14:23

fix(ring-switch): drop dead m-eval helpers

b932849

Remove the unused shifted-eq evaluation helpers and the stale test that only exercised them so CI can keep treating dead code as an error. Made-with: Cursor

cursor bot reviewed Apr 10, 2026

View reviewed changes

src/protocol/sumcheck/hachi_stage1.rs Show resolved Hide resolved

perf(sumcheck): fuse sparse y stage1 handoff

63ed28e

Fuse the sparse y-stage full-table fold with next-round polynomial generation so the y-first Stage 1 cutover recovers the onehot regression without reverting semantics. Made-with: Cursor

quangvdao changed the title ~~feat(sumcheck): implement phase-1 y-first cutover~~ feat(sumcheck): complete y-first cutover and recover perf Apr 10, 2026

quangvdao added 2 commits April 10, 2026 16:11

style: fix rustfmt on advance_stage1_claim generic bound

7350955

Made-with: Cursor

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(sumcheck): complete y-first cutover and recover perf#46

feat(sumcheck): complete y-first cutover and recover perf#46
quangvdao wants to merge 9 commits intomainfrom
quang/shifted-eq-dp

quangvdao commented Apr 7, 2026 •

edited

Loading

Uh oh!

cursor bot commented Apr 7, 2026 •

edited

Loading

Uh oh!

github-actions bot commented Apr 7, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor bot left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

quangvdao commented Apr 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Summary

Key changes

Performance

Test coverage

Files/components impacted

Uh oh!

cursor bot commented Apr 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

Uh oh!

github-actions bot commented Apr 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Report

Single Polynomial x 32 Variables

4 Polynomials x 30 Variables

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

quangvdao commented Apr 7, 2026 •

edited

Loading

cursor bot commented Apr 7, 2026 •

edited

Loading

github-actions bot commented Apr 7, 2026 •

edited

Loading