perf(spectrum): reuse the masked query key in reseed carry loops

[nh13]

The min_intv-truncation carry loops in forward_truncate_below_maximal and span_rc_walk's expand closure re-tokenized the query 32-mer key at every length step (tokenize_32mer(&q[..l]) per depth). But the keyed comparator already masks the key to the active prefix length, so the full-32 key — computed once — is byte-identical at every depth: the bases beyond l are masked off, and for l > 32 the >32bp tail reads the p slice directly. This hoists the key out of both per-depth loops.

Why

Instruction-attributed profiling (perf record -e instructions:u) of hg38 seeding showed tokenize_32mer at ~28% of retired instructions, almost entirely from these per-length-step calls in the two reseed carry loops. (A cycles profile badly understated it at ~11% — the function is pure compute, so cycle-sampling hid it behind memory stalls elsewhere; instruction sampling is the right lens here.)

Impact

Measured on hg38 (t1, load-subtracted, byte-id verified) in a fuller optimization branch: seeding instructions dropped ~28%, and seeding-only wall moved from 0.60× → 0.75× of the FM-index path (t16, load-excluded). This change is the dominant contributor to that delta.

Correctness (byte-identity)

The keyed comparator (compare_query_vs_suffix_2x_keyed_with_mask) masks the supplied key to min(p.len(), 32), so passing the full-32 key with the per-step slice p is provably equivalent to passing tokenize_32mer(&q[..l]). Verified unchanged:

• forward_truncate_below_maximal_equals_oracle
• mem_search_backward_truncated_span_rc_equals_oracle
• keyed_equals_scalar_mode2 and the other keyed comparator proptests

No interface, format, or behavior change — pure compute reduction.

Summary by CodeRabbit

• Refactor
  • Optimized search index operations by improving computational efficiency in boundary detection and prefix matching algorithms. Eliminated redundant per-iteration calculations while preserving identical behavior and accuracy. These changes reduce processing overhead for better performance.

[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: 88c81fb0-0483-47ae-bad6-8a3982d23268

📥 Commits

Reviewing files that changed from the base of the PR and between d2d5c15 and 654c659.

📒 Files selected for processing (1)

• prmi/src/index/spectrum.rs

---

Walkthrough

In spectrum.rs, two sites that called tokenize_32mer on a truncated prefix p/p_key at every iteration now hoist a single q_key = tokenize_32mer(query/q, ...) before the loop and reuse it, relying on the keyed comparator's masking to handle varying length l.

Changes

Key hoisting in RC-walk and forward-truncate loops

Layer / File(s)	Summary
Hoist q_key in expand closure and forward-truncate loop prmi/src/index/spectrum.rs	span_rc_walk's expand closure drops the per-depth p_key = tokenize_32mer(...) and reuses the outer q_key for all shares_prefix calls; forward_truncate_below_maximal hoists q_key once before the while l > 1 loop, eliminating the per-iteration tokenize_32mer(p, ...) in both sites.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• fg-labs/prmi#42: Directly modifies span_rc_walk's outward interval-expansion path, the same closure this PR refactors for key reuse.
• fg-labs/prmi#31: Touches forward_truncate_below_maximal in spectrum.rs, the second function this PR optimizes.
• fg-labs/prmi#27: Introduces the RC-downward decreasing-l outward-scan pattern that both hoisted sites operate within.

🚥 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 captures the main optimization: hoisting query key computation outside reseed carry loops to eliminate redundant tokenization work.
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/v0.3-reseed-key-hoist

---

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.