perf(codegen): inline the slot load for guarded numeric array index reads

[TheHypnoo]

What

For a numeric (raw-f64) array read arr[i] that already passes the
numeric_array_index_get_guard, emit the slot load inline instead of
calling the runtime helper js_array_numeric_get_f64_unboxed.

Why it's safe

The guard on the way into the fast block already proves everything the helper
re-checks:

• a plain, non-forwarded Array (plain_array_index_guard),
• raw-f64 numeric layout (js_array_is_numeric_f64_layout),
• index in bounds (plain_array_index_guard(.., in_bounds=true)).

Raw-f64 arrays are dense (no HOLE slots) and each slot holds a raw f64, so
the helper's hot path reduces to return *elements_ptr.add(index). We emit
exactly that: load double at arr + 8 + i*8 (ArrayHeader: length@0,
capacity@4, elements@8). The non-numeric else branch already addressed
elements the same way.

Impact

array_read micro-benchmark: 149ms → 91ms (~1.6×) by removing the
per-element call + redundant re-validation.

Validation

• 29/29 benchmark-correctness programs match Node byte-for-byte.
• Edge cases match Node: in-bounds sum, OOB index → undefined, negative index
  → undefined.
• Updated tests/typed_feedback.rs to assert the IR no longer contains
  call double @js_array_numeric_get_f64_unboxed and now contains load double.
• Full parity suite: 702 pass / 53 fail, all 53 in the pre-existing
  gap/issue/parity-module baseline (none array-indexing related). Zero new
  regressions.

Summary by CodeRabbit

• Performance

  • Improved the numeric array index “fast path” by eliminating the unboxed helper call and performing an inline double element load in the guarded typed-feedback path.

• Tests

  • Updated typed feedback assertions to confirm the unboxed helper is not invoked and that direct double loads are generated.
  • Updated benchmark IR checks and regression IR fixtures to validate the guarded fast-path block shape and disallow the helper call.

[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: c1a05f9f-a7c3-4d06-a409-459fb9ed64d4

📥 Commits

Reviewing files that changed from the base of the PR and between 73c239d and 966e1aa.

📒 Files selected for processing (1)

• tests/test_compiler_output_regression.py

---

📝 Walkthrough

Walkthrough

In lower_guarded_array_index_get, the require_numeric_layout fast path is changed to load the array element inline: it extends idx_i32 to i64, computes the byte offset (idx*8 + 8), adds that to the masked handle, casts to double*, and loads directly—removing the js_array_numeric_get_f64_unboxed runtime helper call. The corresponding unit test assertion is updated to verify the helper is absent and load double is present. The benchmark IR check and regression test fixture are also updated to assert the new behavior.

Changes

Inline numeric array element load

Layer / File(s)

Summary

Inline load implementation, unit test, and benchmark IR verification crates/perry-codegen/src/expr/index_get.rs, crates/perry-codegen/tests/typed_feedback.rs, benchmarks/compiler_output/workloads.toml, tests/test_compiler_output_regression.py

lower_guarded_array_index_get replaces the js_array_numeric_get_f64_unboxed call with inline pointer arithmetic (index extension → byte offset → masked_handle + offset → double* cast → load double). The typed_feedback_guards_computed_numeric_array_index_hot_path test removes the old positive call assertion, adds a negative assertion for the helper, and adds a positive assertion for load double, with a clarifying comment. The numeric_array_uses_unboxed_get IR check in the workload is updated to match the guarded typed-feedback path and assert the helper is not called. The regression test fixture is updated to reflect the new guard-driven control flow with inlined fast-path computation.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Poem

🐇 No more helper hops across the runtime floor,
The bunny loads the double straight from memory's door.
Extend the index, shift by eight, add header too,
A pointer cast, a load, and we are nearly through!
Inline is faster—that's the carrot I adore. 🥕

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 25.00% which is insufficient. The required threshold is 80.00%.	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 and concisely describes the main optimization: inlining slot loads for guarded numeric array index reads, which is the primary change across all modified files.
Description check	✅ Passed	The PR description includes all key sections: Summary (What/Why), Changes context, detailed validation results, performance metrics, and test updates, comprehensively documenting the optimization and its safety guarantees.
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/inline-numeric-array-get

---

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@benchmarks/compiler_output/workloads.toml`:
- Around line 627-630: The workload check for numeric_array_uses_unboxed_get at
lines 627-630 only verifies the presence of the guard and absence of the helper
function call, but does not assert that the expected direct inline-load
instruction is actually present in the IR output. Add a regex_contains or
contains check to verify the presence of the direct load instruction (such as
"load double") to strengthen the test and ensure it validates the complete
inline-load fast path optimization, preventing regressions where the helper
removal occurs without the expected inline load taking place.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 24f156c4-8dd6-479e-b179-ce281344a06d

📥 Commits

Reviewing files that changed from the base of the PR and between 6c906b0 and 5dfcfd6.

📒 Files selected for processing (1)

• benchmarks/compiler_output/workloads.toml