perf(reseed): neighbour-LCP scan for SMEM truncation (byte-identical, +13.5%)

[nh13]

Summary

Replace the per-length linear length-walk in the reseed truncation primitives — forward_truncate_below_maximal and mem_search_backward_truncated_span_rc (src/index/spectrum.rs) — with an O(min_intv) neighbour-LCP scan. Signatures are unchanged, so the production fused seeding path (collect_smems → these primitives) benefits directly.

Correctness is defined as byte-identity to the existing walk (not a re-derivation of the result): the converged L*/interval and emitted SMEMs are unchanged.

What changed

• forward_truncate_below_maximal: the O(lmax−L*) per-length walk becomes an O(min_intv) scan over neighbouring SA entries' LCPs.
• mem_search_backward_truncated_span_rc: the same neighbour-axis scan applied to the reverse-complement span.

Validation (all green)

• Unit byte-identity: existing oracle proptests (forward_truncate_below_maximal_equals_oracle, mem_search_backward_truncated_span_rc_equals_oracle, span_rc_zero_occ_floors_to_anchor_len) plus a new repeats/ties corpus (24,130 cases incl. shared-L* ties and occ(maximal) ∈ [2, min_intv)).
• Probe budget (the perf change): forward 264 → 18 probes/call on long-unique reads; RC 242 → 15 — new guard tests fail on the old walk, pass on the scan.
• End-to-end SMEM (rid,m,n,occ) parity: 0 differences vs current on E. coli K12 and phiX (fwd+revcomp) across the seeding param grid (msl 17/19/31, max_mem_intv 0, split_width 5 — counts genuinely vary).
• hg38, 200k WGS reads: byte-identical SMEM count vs current.

Performance

Warm hg38, 200k WGS reads, single-thread, load-excluded (the fused collect_smems path): 76,459 → 66,119 ns/read (−13.5%), tight variance, identical SMEM count.

Test plan

• cargo test (oracle + proptests, feature on/off)
• cargo clippy --all-features --all-targets -- -D warnings, cargo fmt --check
• dump 0/0 on ecoli + phiX across the grid
• wall non-regression

Summary by CodeRabbit

• Refactor

  • Optimized internal truncation algorithms for improved performance in candidate discovery.

• Tests

  • Extended test coverage with new correctness and performance regression tests, including edge cases and stress scenarios.

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 4668e4c9-1b30-434e-9146-6ff8097386ce

📥 Commits

Reviewing files that changed from the base of the PR and between 01a2627 and 350b68d.

📒 Files selected for processing (1)

• prmi/src/index/spectrum.rs

---

Walkthrough

Replaces per-base downward walks in span_rc_walk and forward_truncate_below_maximal with neighbor-LCP-driven scans to locate truncation length L*. Adds a structured repeat_corpus() generator and oracle-comparison plus probe-count regression tests for both rewritten paths.

Changes

Neighbor-LCP truncation algorithm and tests

Layer / File(s)	Summary
span_rc_walk and forward_truncate_below_maximal rewrites prmi/src/index/spectrum.rs	span_rc_walk drops the per-base expand(l,…) loop; now caps neighbor-LCPs and extends toward the higher-LCP side until occ >= min_intv. forward_truncate_below_maximal hoists the full-32 q_key, caps neighbor-LCPs at lmax-1, tracks best_len, and terminates on occ >= min_intv or exhausted expansion—replacing the l-decrement/shares_prefix loop.
Repeat corpus and oracle regression tests prmi/src/index/spectrum.rs	repeat_corpus() generates an LCG backbone with embedded motifs, tandem repeats, and homopolymer runs. forward_truncate_equals_oracle_on_repeats_both_modes and span_rc_equals_oracle_on_repeats compare both rewritten paths against brute-force oracles for multiple min_intv values and both want_interval modes; under spectrum-probe-count, probe counts are asserted bounded to guard the new asymptotic.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Possibly related PRs

• fg-labs/prmi#27: Introduces the backward-truncated-span RC primitive using a downward/outward walk—the same span_rc_walk logic this PR replaces with neighbor-LCP scans.
• fg-labs/prmi#31: Adds forward_truncate_below_maximal as a new primitive with the stepwise length-decrement loop that this PR rewrites.
• fg-labs/prmi#45: Modifies the same two functions (span_rc_walk, forward_truncate_below_maximal) to hoist/reuse the full-32 q_key, a precursor change this PR builds on.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the core optimization: replacing per-length walks with neighbour-LCP scans for SMEM truncation, with the quantified performance improvement (+13.5%). It directly reflects the main change.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch perf/reseed-neighbor-scan

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[nh13]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.