docs(plan): record the resolved speed picture

[proggeramlug]

Completes the plan's Part 2 with what today actually established. The earlier update captured the first measurement; four corrections followed, and the corrections are the useful part.

The benchmarks were lying in two different ways

Fixed and shipped in #7403. Discarding the result let Perry eliminate the loop; and even with it consumed, fibonacci(FIB_N) is loop-invariant, so it hoists out and runs once — checksum correct, TOTAL zero.

benchmark	was reported	actually
bench_fibonacci	~240× faster	2.5× faster
bench_bitwise	∞ faster	20.4× SLOWER

The fast path already exists

bench_bitwise's hot loop has no runtime helper calls at all — 4 frem per iteration, 4 × bl _fmod, 1754 profile samples. But expr/binary.rs:588 already emits guarded srem with the IEEE -0 correction. The gap is the analysis, reduced to one line (#7404):

let a = 12345678; … a % 1000            // srem — fires
let a = 12345678; … a % 1000; a = a + 1 // frem — lost

Four wrong hypotheses about that subsystem are recorded so they aren't retried, along with why a careless widening is dangerous: it returns silently wrong integer arithmetic that === tests cannot see.

A measured negative result

js_array_grow's HOLE-init loop is already vectorised (6 stp in the baseline), so the slice::fill rewrite proposed for #7396 measures nothing. The contrary reading came from grepping a symbol absent from a stripped binary — grep -c returns 0 whether or not the pattern exists.

Method notes strengthened

Third instance today of a check that could not fail, and a new one worth its own line: a fix that compiles, passes every correctness test, and changes no emitted instruction is a fix to the wrong site. The guarded srem I wrote for #7404 passed the entire -0 edge-case table and emitted zero srem.

Summary by CodeRabbit

• Documentation
  • Updated performance documentation with corrected benchmark results and validation details.
  • Documented that bench_bitwise is significantly slower than Node due to repeated modulo calculations.
  • Clarified findings around integer analysis, array growth, loop optimization, and store performance.
  • Added notes on rejected optimization hypotheses and improved measurement methodology.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 08b0ca5c-9623-4d10-b52d-e273a263e714

📥 Commits

Reviewing files that changed from the base of the PR and between d5b115d and b7869b4.

📒 Files selected for processing (2)

• changelog.d/7405-plan-speed-final.md
• docs/engine-plan.md

---

📝 Walkthrough

Walkthrough

The PR documents corrected benchmark measurements and expanded validation methods. It attributes the bench_bitwise slowdown to % lowering and integer-valued analysis limits. It also records the rejected js_array_grow rewrite and additional emitted-IR checks.

Changes

Benchmark analysis

Layer / File(s)	Summary
Corrected benchmark findings changelog.d/7405-plan-speed-final.md, docs/engine-plan.md	The documentation adds checksum-based benchmark validation, corrected benchmark ratios, % lowering findings, and the rejected array-growth rewrite.
Expanded validation method docs/engine-plan.md	The method notes add heap-limit, symbol, disassembly, debug-symbol, nm, and emitted-IR validation checks.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related issues

• Issue 7404 — Covers the same bench_bitwise %/fmod bottleneck and integer-analysis limitation.

Possibly related PRs

• PerryTS/perry#6860 — Addresses related integer-value and type analysis.
• PerryTS/perry#6902 — Covers related integer remainder lowering and proof changes.
• PerryTS/perry#7397 — Updates overlapping benchmark methodology and performance findings in docs/engine-plan.md.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch docs/plan-speed-final

---

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.}