[codex] Hoist typed feedback site registration

[andrewtdiz]

Summary

Hoists typed-feedback site registration out of hot use sites and into function-entry setup. The guard, fallback, pass, and counter calls still run at the original use sites, but the per-site metadata registration now happens once per function entry instead of once per loop iteration.

Also makes the benchmark harness usable on this Linux runner by supporting /usr/bin/time -v RSS output and skipping Node columns when the installed Node cannot execute .ts benchmark inputs directly.

Root cause

benchmarks/suite/10_nested_loops.ts LLVM traces showed js_typed_feedback_register_site(...) inside the hot for.body.21 inner loop before each typed-feedback array guard. That turned static site metadata registration into repeated hot-loop work.

Benchmark Results

Local baseline was captured from e816fc3 because benchmarks/baseline.json is stale for this Linux environment.

• ./benchmarks/compare.sh --quick --runs 3 --warn-only --json-out /tmp/perry-final-e816fc3e4.json
• 10_nested_loops: 3383ms -> 956ms, 71.7% faster
• 02_loop_overhead: 74ms -> 74ms
• 05_fibonacci: 262ms -> 261ms
• 06_math_intensive: 70ms -> 69ms
• 13_factorial: 96ms -> 94ms
• ./benchmarks/quick.sh
• 16_matrix_multiply: 6462ms -> 1842ms, 71.5% faster

Node columns/correctness checks were skipped because local Node cannot run .ts benchmark inputs directly.

Validation

• bash -n benchmarks/quick.sh
• bash -n benchmarks/compare.sh
• cargo fmt --check
• git diff --check
• cargo build --release
• cargo test -p perry-codegen --test typed_feedback
• PERRY_BIN=target/release/perry python3 tests/test_typed_feedback_runtime_evidence.py
• tests/test_benchmark_output_verifier.sh
• target/release/perry compile --no-cache benchmarks/suite/10_nested_loops.ts -o /tmp/perry-nested-loops-final --trace llvm --quiet
• confirmed registration calls are in entry setup only and not in for.body.21
• /tmp/perry-nested-loops-final produced nested_loops:963 and sum:26991000000
• target/release/perry compile --no-cache benchmarks/suite/16_matrix_multiply.ts -o /tmp/perry-matrix-multiply-final --quiet && /tmp/perry-matrix-multiply-final
• produced matrix_multiply:1778 and checksum:41079519680

Summary by CodeRabbit

• Performance

  • Optimized compiler's typed feedback registration mechanism for improved benchmark performance.

• Documentation

  • Added performance run documentation recording baseline and post-optimization benchmark results.

• Chores

  • Enhanced benchmark scripts with improved TypeScript environment detection and refined peak memory measurement across operating systems.

[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: 2f3898f5-293e-44eb-b587-715aec423adf

📥 Commits

Reviewing files that changed from the base of the PR and between 0f73bb0 and 00057f5.

📒 Files selected for processing (5)

• PERF_RUN_LOG.md
• benchmarks/compare.sh
• benchmarks/quick.sh
• crates/perry-codegen/src/expr/typed_feedback.rs
• crates/perry-codegen/src/function.rs

---

📝 Walkthrough

Walkthrough

Moves js_typed_feedback_register_site calls from per-guard-use-site block emission to a single function-entry setup region via a new LlFunction::entry_setup_call_void helper. Benchmark scripts gain cross-platform Node TypeScript runner detection, awk-based timing/RSS parsing, and consistent NODE_CMD wiring. A new perf log entry records the measured results.

Changes

Typed Feedback Registration Hoist

Layer / File(s)	Summary
Function entry setup call helper crates/perry-codegen/src/function.rs	Clarifies entry_post_init_setup placement docs and adds LlFunction::entry_setup_call_void to record FFI calls, format a typed LLVM call void IR line, and append it to the entry setup region.
Typed feedback registration emission crates/perry-codegen/src/expr/typed_feedback.rs	emit_typed_feedback_register_site precomputes site_id, kind, and *_len argument strings then emits registration via ctx.func.entry_setup_call_void instead of ctx.block().call_void.

Benchmark Harness Runner and RSS Updates

Layer / File(s)	Summary
Node TS detection and cross-platform RSS measurement benchmarks/compare.sh, benchmarks/quick.sh	Both scripts add detect_node_ts_runner to probe node vs node --experimental-strip-types, rewrite timing extraction with awk, and refactor RSS collection to use OS-specific /usr/bin/time flags with awk-based peak RSS parsing for macOS and Linux.
NODE_CMD wiring in measurement and results loop benchmarks/compare.sh, benchmarks/quick.sh	Node RSS measurement in compare.sh uses ${NODE_CMD[@]}; quick.sh replaces `cut -d'
Performance run documentation PERF_RUN_LOG.md	Adds the 2026-06-17 run entry with baseline and post-change benchmark commands, results for nested_loops and matrix_multiply, verification steps, and a PR link.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• PerryTS/perry#5084: Modifies the same emit_typed_feedback_register_site function in crates/perry-codegen/src/expr/typed_feedback.rs, controlling whether the registration call is emitted at all — directly upstream of the hoisting change in this PR.

Poem

🐇 Hop hop, no more per-guard repeat,
One call at function entry — neat!
The benchmarks run on Linux now,
With awk to parse the peak RSS row.
The loops and matrices go zoom,
And Perry's codegen finds more room! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and specifically describes the main optimization: hoisting typed feedback site registration from hot paths to function entry.
Description check	✅ Passed	The description covers summary, root cause, benchmark results, and comprehensive validation steps, but is missing required template sections like explicit 'Changes' bullets and test plan checkboxes.
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 codex/perry-performance-20260617

---

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