perf(#361): skip realm-proto wrappers in call_value for same-realm callees by dowdiness · Pull Request #368 · dowdiness/js_engine

dowdiness · 2026-06-16T09:56:08Z

Summary

Adds a fast path in `Interpreter::call_value` that skips both `with_active_value` + `with_active_callee_realm_value` wrappers when no active-override slot is set and the callee belongs to the main realm
Packs all 10 realm proto slots into a single `Array[Value]` entry at symbol key `-101`, reducing the fast-path check from 10 HashMap lookups to 1 lookup + 10 sequential array reads

Motivation

The two realm-proto wrapper layers in `call_value` perform ~80 Ref ops + 10 HashMap lookups on every function call to save/apply/restore 10 prototype override slots. In single-realm bytecode execution (the common case), both wrappers are pure overhead.

Fast path correctness

Two parts, both required:

Part 1 — all 10 `active_*_prototype_override` Refs are `None`. Checked individually; sampling a subset is unsafe because the `RealmState` Refs are public.
Part 2 — every callee stamped proto slot is absent (Null in the packed array) or pointer-equals the main realm's corresponding proto. All 10 slots are checked; `stamp_function_realm_with` accepts them independently, so sampling one slot is unsafe.

When both hold, the wrappers produce identical results to skipping them.

Two whitebox positive-control tests in `factories_wbtest.mbt` verify scenarios the old single-proto check would have passed incorrectly:

Active `array_prototype_override` set while `function` slot is None
Callee stamped with foreign `Array.prototype` while `function`/`object` slots match the realm

Storage change

Realm protos were stored as 10 separate entries in `symbol_properties` (keys -101..-110). They are now packed into a single 10-element `Array[Value]` at key -101 (`FUNCTION_REALM_PROTOS_PACKED_SYMBOL_ID`). Keys -102..-110 are freed. The fast path reads the array elements directly (sequential memory, warm cache line) rather than doing 10 independent HashMap lookups.

Benchmark (10-run median, JS target, vs post-#367 baseline)

benchmark	before	after	delta
`isolate/bytecode/call_frame`	7.514 ms	6.295 ms	−16%
`isolate/bytecode/method_call`	8.562 ms	7.086 ms	−17%
`isolate/bytecode/runtime_helpers`	10.369 ms	9.013 ms	−13%

Test plan

`moon test` passes 2096/2096 (including 2 new positive-control fast-path whitebox tests)
`moon check --deny-warn` clean
CI test262 (cross-realm regression gate)

🤖 Generated with Claude Code

coderabbitai · 2026-06-16T09:56:21Z

📝 Walkthrough

Walkthrough

Adds a private is_cross_realm_callee helper in factories.mbt that detects cross-realm function objects via a realm-stamp slot and proxy-following recursion. call_value in call.mbt uses this helper to skip the with_active_value/with_active_callee_realm_value wrappers when no prototype override is active and the callee is same-realm.

Changes

Cross-realm callee fast path

Layer / File(s)	Summary
`is_cross_realm_callee` helper `interpreter/runtime/factories.mbt`	New private function reads `FUNCTION_REALM_FUNCTION_PROTO_SYMBOL_ID` from the callee, follows proxy targets recursively, and returns `false` for non-function or uninitialized callees.
Fast path in `call_value` `interpreter/runtime/call.mbt`	`call_value` now calls `call_value_impl` directly when `active_function_prototype_override` is absent and the callee is not cross-realm; the existing realm-wrapper path is kept for all other cases.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐇 Hop, hop, skip the wrapper wide,
When realms align, no need to hide!
A stamp on proto tells the tale—
Same realm? Fast path! No need to flail.
The rabbit zips through, quick as a blink,
Less wrapping means more time to think! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly describes the main optimization: skipping realm-proto wrappers for same-realm callees, which directly matches the PR's primary change and performance objective.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

_{✏️ 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/call-value-realm-fast-path

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

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

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 57d70339b3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

github-actions · 2026-06-16T10:02:47Z

Benchmark Results

Run: https://github.com/dowdiness/js_engine/actions/runs/27615428559

startup/tiny_program is the PR #153 / issue #141 guardrail for built-in realm-stamping startup cost.

Stage summary

stage	benchmarks	total mean	slowest benchmark	slowest mean	noisy rows
startup	3	2.282 ms	startup/tiny_program	1.158 ms	0
frontend	7	0.871 ms	pipeline/parse_heavy	0.497 ms	2
execution	25	14313.208 ms	exec/fibonacci_30	13020.802 ms	2

Focused bytecode base-vs-head comparison

Base-vs-head deltas are reporting-only. Negative delta and PR/base < 1.00x mean the PR is faster; interpret high-CV or noisy rows cautiously.

benchmark	stage	base mean	PR mean	delta	PR/base	base CV	PR CV	noisy
baseline/bytecode/closure_factory	execution	14.931 ms	12.643 ms	-15.3%	0.85x	5.3%	4.3%	no
pipeline/bytecode/evaluate	execution	9.956 ms	9.211 ms	-7.5%	0.93x	3.3%	2.6%	no
isolate/bytecode/call_frame	execution	10.351 ms	8.541 ms	-17.5%	0.83x	0.8%	4.7%	no
isolate/bytecode/runtime_helpers	execution	13.863 ms	12.221 ms	-11.8%	0.88x	0.9%	1.2%	no
isolate/bytecode/local_access	execution	36.250 ms	36.966 ms	+2.0%	1.02x	1.2%	2.3%	no
isolate/bytecode/env_access	execution	39.620 ms	36.754 ms	-7.2%	0.93x	2.2%	1.9%	no
isolate/bytecode/captured_access	execution	35.661 ms	37.575 ms	+5.4%	1.05x	1.3%	1.5%	no
isolate/bytecode/dispatch_stack	execution	22.649 ms	23.935 ms	+5.7%	1.06x	1.0%	2.6%	no

Base-vs-head comparison

benchmark	stage	base mean	PR mean	delta	PR/base	base CV	PR CV	noisy
startup/tiny_program	startup	1.415 ms	1.158 ms	-18.2%	0.82x	5.5%	6.7%	no
lexer/small	frontend	0.031 ms	0.030 ms	-3.8%	0.96x	23.2%	19.3%	base, PR
lexer/large	frontend	0.266 ms	0.264 ms	-0.6%	0.99x	3.7%	7.9%	no
exec/fibonacci_30	execution	15188.602 ms	13020.802 ms	-14.3%	0.86x	0.3%	0.7%	no
exec/property_chain	execution	13.739 ms	13.616 ms	-0.9%	0.99x	7.5%	14.6%	no
startup/phase/parse_tiny	frontend	0.002 ms	0.002 ms	+7.7%	1.08x	0.8%	0.6%	no
startup/phase/new_interpreter	startup	1.390 ms	1.124 ms	-19.1%	0.81x	11.9%	13.9%	no
startup/phase/execute_preparsed_tiny	execution	0.000 ms	0.000 ms	+4.8%	1.05x	2.1%	0.8%	no
startup/phase/event_loop_drain_empty	startup	0.000 ms	0.000 ms	-3.9%	0.96x	0.7%	0.6%	no
startup/phase/result_stringify_output	execution	0.000 ms	0.000 ms	+0.3%	1.00x	0.5%	0.5%	no
exec/array_map_filter	execution	21.227 ms	19.839 ms	-6.5%	0.93x	17.2%	21.6%	base, PR
exec/closure_factory	execution	30.885 ms	29.701 ms	-3.8%	0.96x	5.3%	4.8%	no
baseline/closure_legacy/closure_factory	execution	28.957 ms	28.148 ms	-2.8%	0.97x	9.4%	9.6%	no
baseline/bytecode/closure_factory	execution	14.931 ms	12.643 ms	-15.3%	0.85x	5.3%	4.3%	no
isolate/bytecode/dispatch_stack	execution	22.649 ms	23.935 ms	+5.7%	1.06x	1.0%	2.6%	no
isolate/bytecode/local_access	execution	36.250 ms	36.966 ms	+2.0%	1.02x	1.2%	2.3%	no
isolate/bytecode/env_access	execution	39.620 ms	36.754 ms	-7.2%	0.93x	2.2%	1.9%	no
isolate/bytecode/captured_access	execution	35.661 ms	37.575 ms	+5.4%	1.05x	1.3%	1.5%	no
isolate/bytecode/call_frame	execution	10.351 ms	8.541 ms	-17.5%	0.83x	0.8%	4.7%	no
isolate/bytecode/runtime_helpers	execution	13.863 ms	12.221 ms	-11.8%	0.88x	0.9%	1.2%	no
isolate/bytecode/property_get	execution	45.654 ms	44.657 ms	-2.2%	0.98x	1.6%	3.8%	no
isolate/bytecode/property_set	execution	42.442 ms	40.842 ms	-3.8%	0.96x	2.3%	0.9%	no
isolate/bytecode/method_call	execution	11.356 ms	9.429 ms	-17.0%	0.83x	3.8%	0.5%	no
isolate/bytecode/object_literal	execution	14.073 ms	13.229 ms	-6.0%	0.94x	0.7%	1.6%	no
isolate/bytecode/array_literal	execution	14.648 ms	14.326 ms	-2.2%	0.98x	0.7%	1.6%	no
exec/arithmetic_loop	execution	899.341 ms	841.581 ms	-6.4%	0.94x	0.4%	0.6%	no
exec/object_construction	execution	7.446 ms	7.084 ms	-4.9%	0.95x	5.7%	6.3%	no
exec/string_ops	execution	2.137 ms	1.884 ms	-11.8%	0.88x	17.0%	16.8%	base, PR
pipeline/exec/lex	frontend	0.028 ms	0.027 ms	-1.8%	0.98x	2.4%	0.5%	no
pipeline/exec/parse	frontend	0.027 ms	0.028 ms	+3.7%	1.04x	3.2%	3.2%	no
pipeline/exec/evaluate	execution	25.857 ms	25.658 ms	-0.8%	0.99x	4.5%	12.5%	no
pipeline/closure_legacy/evaluate	execution	26.334 ms	24.564 ms	-6.7%	0.93x	4.8%	4.5%	no
pipeline/bytecode/compile	frontend	0.022 ms	0.023 ms	+4.0%	1.04x	28.2%	23.2%	base, PR
pipeline/bytecode/evaluate	execution	9.956 ms	9.211 ms	-7.5%	0.93x	3.3%	2.6%	no
pipeline/parse_heavy	frontend	0.490 ms	0.497 ms	+1.5%	1.02x	5.2%	8.6%	no

Mean-time chart (log scale)

benchmark	stage	mean	chart
startup/tiny_program	startup	1.158 ms	`##`
lexer/small	frontend	0.030 ms ⚠	`#`
lexer/large	frontend	0.264 ms	`#`
exec/fibonacci_30	execution	13020.802 ms	`##############################`
exec/property_chain	execution	13.616 ms	`########`
startup/phase/parse_tiny	frontend	0.002 ms	`#`
startup/phase/new_interpreter	startup	1.124 ms	`##`
startup/phase/execute_preparsed_tiny	execution	0.000 ms	`#`
startup/phase/event_loop_drain_empty	startup	0.000 ms	`#`
startup/phase/result_stringify_output	execution	0.000 ms	`#`
exec/array_map_filter	execution	19.839 ms ⚠	`#########`
exec/closure_factory	execution	29.701 ms	`##########`
baseline/closure_legacy/closure_factory	execution	28.148 ms	`##########`
baseline/bytecode/closure_factory	execution	12.643 ms	`########`
isolate/bytecode/dispatch_stack	execution	23.935 ms	`##########`
isolate/bytecode/local_access	execution	36.966 ms	`###########`
isolate/bytecode/env_access	execution	36.754 ms	`###########`
isolate/bytecode/captured_access	execution	37.575 ms	`###########`
isolate/bytecode/call_frame	execution	8.541 ms	`#######`
isolate/bytecode/runtime_helpers	execution	12.221 ms	`########`
isolate/bytecode/property_get	execution	44.657 ms	`############`
isolate/bytecode/property_set	execution	40.842 ms	`###########`
isolate/bytecode/method_call	execution	9.429 ms	`#######`
isolate/bytecode/object_literal	execution	13.229 ms	`########`
isolate/bytecode/array_literal	execution	14.326 ms	`########`
exec/arithmetic_loop	execution	841.581 ms	`#####################`
exec/object_construction	execution	7.084 ms	`######`
exec/string_ops	execution	1.884 ms ⚠	`###`
pipeline/exec/lex	frontend	0.027 ms	`#`
pipeline/exec/parse	frontend	0.028 ms	`#`
pipeline/exec/evaluate	execution	25.658 ms	`##########`
pipeline/closure_legacy/evaluate	execution	24.564 ms	`##########`
pipeline/bytecode/compile	frontend	0.023 ms ⚠	`#`
pipeline/bytecode/evaluate	execution	9.211 ms	`#######`
pipeline/parse_heavy	frontend	0.497 ms	`#`

Closure-conversion comparison

unavailable

…llees Add a fast path in Interpreter::call_value that bypasses both with_active_value + with_active_callee_realm_value when every active- override slot is None and every callee stamped proto matches the main realm's (or is absent). Storage: pack all 10 realm proto slots into a single 10-element Array[Value] at symbol key -101 (FUNCTION_REALM_PROTOS_PACKED_SYMBOL_ID) instead of 10 separate symbol_properties entries. The fast path check does 1 HashMap lookup + 10 sequential array reads + 10 physical_equal comparisons vs 10 HashMap lookups in the earlier full-10-slot approach. Keys -102..-110 are freed. Safety: all 10 active-override Ref slots are checked (Part 1) and all 10 callee stamped proto slots are compared (Part 2), so the check is safe even when stamp_function_realm_with sets slots independently or when a non-function override is active. Two whitebox positive-control tests verify that the OLD single-proto check would have failed both scenarios (active_array_prototype_override set, mixed-stamp with foreign array proto) while the full check correctly blocks the fast path. Benchmark improvement vs post-#367 baseline (10-run median): isolate/bytecode/call_frame: 7.514 → 6.295 ms (−16%) isolate/bytecode/method_call: 8.562 → 7.086 ms (−17%) isolate/bytecode/runtime_helpers: 10.369 → 9.013 ms (−13%) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…rotos?] (#369) (#369) * perf: replace 10 Ref reads in realm_fast_path_allowed with single Bool Replace the 10 double-indirect Ref reads in Part 1 of realm_fast_path_allowed with a single Bool field (has_active_override) maintained by apply_active_realm_protos. The Bool is set to true when any active override is Some, false when all are None. Ablation established that the 10 reads + HashMap lookup consumed 14-22% of per-call time (PR #367/#368 session). This change eliminates Part 1 (10 x 2 pointer dereferences) at the cost of 1 direct field read. Measured gain on JS target (5-run median): call_frame 6.66 ms → 6.25 ms (−6.2%, CV 4.7% → 0.9%) method_call 7.66 ms → 7.01 ms (−8.5%) runtime_helpers 9.49 ms → 8.79 ms (−7.4%) local_access (control): noise only All 7 raw direct-write sites in wbtest files updated to maintain the invariant (has_active_override == OR of all 10 active_*_override Refs). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * fix: update remaining has_active_override write sites in has_property_realm_wbtest Three direct writes to active_{map,set,promise}_prototype_override.val in has_property_realm_wbtest.mbt were missing the companion has_active_override = true update. All 10 direct raw write sites in the source tree are now audited and consistent with the invariant: has_active_override == (any active_*_prototype_override.val is Some). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * fix(perf): replace 10-Ref + Bool cache with Ref[FunctionRealmProtos?] PR #369 introduced a stale-cache hazard: the 10 active_*_prototype_override Ref fields remained publicly writable (RealmState is pub(all)), so external code writing any Ref directly would bypass has_active_override and allow call_value to take the fast path incorrectly. Fix: collapse the 10 individual Refs + Bool into a single Ref[FunctionRealmProtos?] on RealmState. - None = no cross-realm context active - Some(protos) = at least one override set The "any active?" check in realm_fast_path_allowed Part 1 is now `active_overrides.val is None` — one Ref read. This is structurally correct regardless of any external write, because writing to the single combined Ref atomically updates both the proto data AND the "any active?" check. No separate Bool cache exists to become stale. Public API changes: - Remove: 10 active_*_prototype_override fields, has_active_override Bool - Add: active_overrides : Ref[FunctionRealmProtos?] - Promote to pub: apply_active_realm_protos, FunctionRealmProtos struct + constructor All wbtest direct-write sites updated to use apply_active_realm_protos. apply_active_realm_protos is now simpler: 10 Ref writes → 1 Option write. active_realm_protos is now simpler: 10 .val reads → unwrap one Option. 2096/2096 tests green. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

chatgpt-codex-connector Bot reviewed Jun 16, 2026

View reviewed changes

Comment thread interpreter/runtime/call.mbt Outdated

dowdiness force-pushed the perf/call-value-realm-fast-path branch from 57d7033 to 96355bb Compare June 16, 2026 11:50

dowdiness merged commit 722db05 into main Jun 16, 2026
15 checks passed

dowdiness deleted the perf/call-value-realm-fast-path branch June 16, 2026 12:36

dowdiness mentioned this pull request Jun 16, 2026

perf: collapse 10 active-override Refs into single Ref[FunctionRealmProtos?] (#369) #369

Merged

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf(#361): skip realm-proto wrappers in call_value for same-realm callees#368

perf(#361): skip realm-proto wrappers in call_value for same-realm callees#368
dowdiness merged 1 commit into
mainfrom
perf/call-value-realm-fast-path

dowdiness commented Jun 16, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Jun 16, 2026 •

edited

Loading

Walkthrough

Changes

Estimated code review effort

Poem

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

github-actions Bot commented Jun 16, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

dowdiness commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Motivation

Fast path correctness

Storage change

Benchmark (10-run median, JS target, vs post-#367 baseline)

Test plan

Uh oh!

coderabbitai Bot commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Estimated code review effort

Poem

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

github-actions Bot commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Results

Stage summary

Focused bytecode base-vs-head comparison

Base-vs-head comparison

Mean-time chart (log scale)

Closure-conversion comparison

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

dowdiness commented Jun 16, 2026 •

edited

Loading

coderabbitai Bot commented Jun 16, 2026 •

edited

Loading

github-actions Bot commented Jun 16, 2026 •

edited

Loading