perf(codegen): retain guarded array accumulators as i32

[proggeramlug]

Summary

• propagate runtime-backed number[] parameter proofs into native-region fact collection
• retain guarded-array Feistel accumulators in canonical i32 slots after a conservative root-level normalization proof
• keep nested bitwise operands native across mutable numeric locals and bounds-proven typed-array reads
• preserve the generic fallback when the entry guard or dominance proof is absent

typed_array result

Exact issue workload, three-run medians on this host:

metric	issue baseline	this PR	change
retired instructions	14.68B	9.66B	-34.2%
cycles	8.10B	5.96B	-26.5%
user CPU	2.73s	2.01s	-26.4%

The specialized encipherUntyped function contains zero js_number_coerce calls after this change. Output remains checksum:-821955270.

Validation

• cargo test -p perry-codegen --lib (1,163 passed, 1 ignored)
• cargo test --release -p perry-codegen --lib (1,163 passed, 1 ignored)
• cargo clippy -p perry --bins
• cargo check -p perry
• release build of perry, perry-runtime-static, and perry-stdlib-static
• exact workload checksum, symbol inspection, and three-run hardware-counter comparison

No version bump.

Fixes #8606

Summary by CodeRabbit

• Performance Improvements
  • Improved numeric-array and typed-array processing for bitwise operations.
  • Reduced unnecessary numeric conversions in optimized execution paths.
  • Preserved integer representations more effectively for faster calculations.
  • Maintained existing results, including checksum consistency.
• Bug Fixes
  • Improved handling of nested bitwise operations, conditional paths, and out-of-bounds array reads.
  • Ensured numeric-array optimizations apply only when type and normalization guarantees are valid.
• Tests
  • Added coverage for guarded numeric arrays, normalization order, and unreachable execution paths.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b7721a4f-20db-45de-b694-e1fecefcae9d

📥 Commits

Reviewing files that changed from the base of the PR and between 5f2c6c6 and 659d6ee.

📒 Files selected for processing (7)

• changelog.d/8613-guarded-array-i32.md
• crates/perry-codegen/src/codegen/function.rs
• crates/perry-codegen/src/codegen/ordinary_param_guard_tests.rs
• crates/perry-codegen/src/collectors/hir_facts.rs
• crates/perry-codegen/src/collectors/int_valued_ta_locals.rs
• crates/perry-codegen/src/collectors/int_valued_ta_locals/tests.rs
• crates/perry-codegen/src/expr/mod.rs

Included review availability: Your plan provides up to 8 included reviews per hour; 5 remain after this review.

---

📝 Walkthrough

Walkthrough

The change propagates numeric-array parameter proofs into native fact collection. Guarded numeric-array reads can seed exact i32 analysis after valid root-level normalization. Bitwise lowering consumes the proof, and tests verify canonical i32 representation without js_number_coerce.

Changes

Guarded numeric-array i32 optimization

Layer / File(s)	Summary
Numeric-array specialization wiring crates/perry-codegen/src/codegen/function.rs, crates/perry-codegen/src/collectors/hir_facts.rs	Specialized parameters with Number or Int32 array elements are collected and passed through native fact-graph construction.
Guarded-array fact and observation analysis crates/perry-codegen/src/collectors/int_valued_ta_locals.rs, crates/perry-codegen/src/collectors/int_valued_ta_locals/tests.rs	The collector recognizes guarded numeric-array reads, preserves OOB undefined behavior, tracks exact i32 normalization, and validates control-flow cases with tests.
Bitwise lowering and specialization validation crates/perry-codegen/src/expr/mod.rs, crates/perry-codegen/src/codegen/ordinary_param_guard_tests.rs, changelog.d/8613-guarded-array-i32.md	Bitwise operands use region-proven i32 values or finite-aware ToInt32 conversion. The code-generation test checks that the specialized clone has no js_number_coerce calls. The changelog records measured performance results.

Estimated code review effort: 4 (Complex) | ~45 minutes

Merge Risk: ⚪ Minimal · up to 659d6

This change specializes guarded numeric array code generation to retain i32 values and reports substantial benchmark improvements while preserving the generic fallback; no actionable merge-blocking risk remains after normal checks and review.

Sequence Diagram(s)

sequenceDiagram
  participant FunctionSpecialization
  participant NativeFactGraph
  participant I32LocalCollector
  participant BitwiseLowering
  FunctionSpecialization->>NativeFactGraph: pass numeric-array parameter proofs
  NativeFactGraph->>I32LocalCollector: pass guarded numeric-array IDs
  I32LocalCollector->>I32LocalCollector: classify reads and root-level exact normalization
  I32LocalCollector->>BitwiseLowering: provide region-proven i32 expressions
  BitwiseLowering->>BitwiseLowering: lower bitwise operands without js_number_coerce

Loading

Suggested reviewers: jdalton

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 68.57% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 35 functions across 6 files. (1 skipped: 1 unsupported.)	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly describes the primary codegen optimization: retaining guarded array accumulators as i32.
Description check	✅ Passed	The description explains the changes, links issue #8606, reports performance results, and lists validation commands.
Linked Issues check	✅ Passed	The changes address #8606 by optimizing the typed_array path, preserving the required checksum, and reducing retired instructions and execution time.
Out of Scope Changes check	✅ Passed	The code, tests, and changelog entry are directly related to the guarded-array codegen optimization and its typed_array performance target.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[proggeramlug]

Merging as a validated batch of three, stacked on current main.

check	result
cargo check --workspace --all-targets	exit 0
perry-codegen --lib	1167 passed, 1 ignored
perry-runtime --lib (RUST_TEST_THREADS=1)	2634 passed, 4 ignored
all seven ratchets	0
cargo fmt --all -- --check	0

Ratchets re-run against the current baseline immediately before merge.

#8610's ratchet debt is resolved. It was held twice for +2 bare raw-handle reads in class_registry/construct.rs (7 against a ceiling of 5) and +1 open-coded StringHeader payload offset. On head c9b994d0f both gates now report clean — 925/925 raw-handle reads, 108 modules within ceilings, payload inventory exit 0. Converted rather than baseline-refreshed, which was the right call given the counts had gone up. Thanks for taking that rather than reaching for --write-baseline; construct.rs runs during instance construction, so a raw handle held across an allocating call there is exactly the shape that yields a stale pointer surfacing cycles later somewhere unrelated.

#8613 and #8614 are aimed at the two tickets filed after the last sweep — #8606 (typed_array round 2, where the wall ratio now exceeds the instruction ratio) and #8607 (pipeline, still 1.74x instructions at ~1.05x cycles). Worth being explicit about what this validation does and does not establish: the gates and suites say these are safe to land, not that they are faster. Both are perf changes, and the corpus sweep after merging is the actual verdict. I will run it and report against those tickets.

Two metadata fixes applied while staging (fork PRs, so they could not be pushed to the branches): #8610 shipped without a changelog.d fragment, so I wrote one; #8614's was keyed 8607- after its issue and is now PR-keyed.