perf(allocator): store pointers directly in `Arena` by overlookmotel · Pull Request #21483 · oxc-project/oxc

overlookmotel · 2026-04-15T21:37:21Z

Arena stores pointers to current bump cursor, and start of the chunk in ChunkFooter.

This means that every allocation involves pointer chasing - read the pointer to theChunkFooter, then read the ChunkFooter's cursor_ptr and start_ptr fields. Because the pointer to the ChunkFooter is wrapped in a Cell, compiler likely cannot assume the value is still what it was last time it read the field, and will read it over and over repeatedly. This adds ~4 cycles of latency to every allocation.

Instead, store these pointers as fields of the Arena itself, to avoid this indirection.

start_ptr still also needs to be stored in ChunkFooter for use when deallocating chunks. And cursor_ptr is also stored in ChunkFooter to support iter_allocated_chunks and iter_allocated_chunks_raw methods.

0.3% - 0.5% perf improvement in parser benchmarks. Allocation is so fast already, that the impact is small - allocation is not the bottleneck. But a micro-benchmark testing allocation in isolation shows that allocation itself gets a 2x speed-up.

See comment on Arena for details of a field layout oddity which has a huge impact on aarch64 (Apple Silicon).

overlookmotel · 2026-04-15T21:37:35Z

perf(allocator): store pointers directly in Arena #21483 👈 (View in Graphite)
main

How to use the Graphite Merge Queue

Add either label to this PR to merge it via the merge queue:

0-merge - adds this PR to the back of the merge queue
hotfix - for urgent changes, fast-track this PR to the front of the merge queue

You must have a Graphite account in order to use the merge queue. Sign up using this link.

_{An organization admin has enabled the Graphite Merge Queue in this repository.} _{Please do not merge from GitHub as this will restart CI on PRs being processed by the merge queue.}

This stack of pull requests is managed by Graphite. Learn more about stacking.

codspeed-hq · 2026-04-15T21:43:40Z

Merging this PR will not alter performance

✅ 48 untouched benchmarks
⏩ 3 skipped benchmarks¹

_{Comparing om/04-15-perf_allocator_store_pointers_directly_in_arena_ (421a549) with main (9c9b6a2)²}

3 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩
No successful run was found on main (6c8684b) during the generation of this report, so 9c9b6a2 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

Copilot

Pull request overview

This PR optimizes oxc_allocator::arena::Arena’s fast-path allocations by storing the current chunk’s bump cursor and start pointer directly on Arena, avoiding repeated indirection through the ChunkFooter (and repeated loads through a Cell).

Changes:

Added Arena::cursor_ptr and Arena::start_ptr fields (both Cell<NonNull<u8>>) and initialized/maintained them across arena lifecycle operations.
Repurposed ChunkFooter::cursor_ptr to store the final cursor only for retired chunks (to support chunk iteration APIs).
Updated chunk iteration to source the current chunk’s cursor from Arena while continuing to read retired chunk cursors from each footer.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
crates/oxc_allocator/src/arena/mod.rs	Adds `cursor_ptr`/`start_ptr` to `Arena`, adjusts `ChunkFooter` fields, removes `ChunkFooter::as_raw_parts`.
crates/oxc_allocator/src/arena/from_raw_parts.rs	Updates raw-transfer construction and cursor mutation to use `Arena::cursor_ptr`.
crates/oxc_allocator/src/arena/drop.rs	Updates `reset()` to restore the arena-level cursor pointer.
crates/oxc_allocator/src/arena/create.rs	Initializes arena-level `start_ptr`/`cursor_ptr` in `new_impl`, updates footer construction.
crates/oxc_allocator/src/arena/chunks.rs	Uses arena-level cursor for current chunk iteration; stores it in the iterator for the first step.
crates/oxc_allocator/src/arena/alloc_impl.rs	Moves fast-path allocation reads/writes to arena-level pointers; syncs retiring chunk cursor back into its footer.

Copilot

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 4 comments.

Copilot

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

overlookmotel · 2026-04-16T14:09:13Z

Merge activity

Apr 16, 2:09 PM UTC: The merge label '0-merge' was detected. This PR will be added to the Graphite merge queue once it meets the requirements.
Apr 16, 2:09 PM UTC: overlookmotel added this pull request to the Graphite merge queue.
Apr 16, 2:15 PM UTC: Merged by the Graphite merge queue.

`Arena` stores pointers to current bump cursor, and start of the chunk in `ChunkFooter`. This means that every allocation involves pointer chasing - read the pointer to the`ChunkFooter`, then read the `ChunkFooter`'s `cursor_ptr` and `start_ptr` fields. Because the pointer to the `ChunkFooter` is wrapped in a `Cell`, compiler likely cannot assume the value is still what it was last time it read the field, and will read it over and over repeatedly. This adds ~4 cycles of latency to every allocation. Instead, store these pointers as fields of the `Arena` itself, to avoid this indirection. `start_ptr` still also needs to be stored in `ChunkFooter` for use when deallocating chunks. And `cursor_ptr` is also stored in `ChunkFooter` to support `iter_allocated_chunks` and `iter_allocated_chunks_raw` methods. 0.3% - 0.5% perf improvement in parser benchmarks. Allocation is so fast already, that the impact is small - allocation is not the bottleneck. But a micro-benchmark testing allocation in isolation shows that allocation itself gets a 2x speed-up. See comment on `Arena` for details of a field layout oddity which has a huge impact on aarch64 (Apple Silicon).

### 🐛 Bug Fixes - 48967e8 isolated_declarations: Drop required type check for private parameter properties on private constructors (#21515) (Dunqing) - 91e5bde transformer/typescript: Preserve computed-key static block when class has an empty constructor (#21562) (Dunqing) - 50e9d26 mangler: Assign correct slot to shadowed function-expression names (#21535) (Dunqing) - 065ce47 isolated_declarations: Collect types from private accessors for paired inference (#21516) (Dunqing) - 00fc136 codegen: Preserve coverage comments before object properties (#21312) (bab) - d676e0c minifier: Mark LHS of `??=` as read when converting from `== null &&` (#21546) (Gunnlaugur Thor Briem) ### ⚡ Performance - e45efc5 parser: Reduce `try_parse` usage in favour of `lookahead` (#21532) (Boshen) - ddb1bf8 parser: Avoid redundant `IdentifierReference` clone in shorthand property (#21511) (Boshen) - be2b392 allocator: Store pointers directly in `Arena` (#21483) (overlookmotel) Co-authored-by: Boshen <1430279+Boshen@users.noreply.github.com> Co-authored-by: Cameron <cameron.clark@hey.com>

The chunks in an `Arena` form a linked list. Previously the list was terminated with a canonical empty chunk, defined as a `static`. Now that we store `start_ptr` and `cursor_ptr` in `Arena` itself (#21483), an empty `Arena` doesn't need to have a pointer to a chunk. Remove the empty chunk, and use `None` to signify the end of the linked list instead. This makes checking "is this chunk the last?" a little cheaper (comparison to 0, not a 64-bit static value), and feels less hacky, and more explicit. It also removes the potential hazard of accidentally mutating the immutable `static` empty chunk.

github-actions Bot added A-allocator Area - Allocator C-performance Category - Solution not expected to change functional behavior, only performance labels Apr 15, 2026

overlookmotel self-assigned this Apr 15, 2026

overlookmotel marked this pull request as ready for review April 15, 2026 21:51

Copilot AI review requested due to automatic review settings April 15, 2026 21:51

Copilot started reviewing on behalf of overlookmotel April 15, 2026 21:52 View session

Copilot AI reviewed Apr 15, 2026

View reviewed changes

overlookmotel force-pushed the om/04-15-perf_allocator_store_pointers_directly_in_arena_ branch from 8dae04d to 2df9f19 Compare April 15, 2026 22:11

overlookmotel marked this pull request as draft April 15, 2026 22:44

overlookmotel force-pushed the om/04-15-perf_allocator_store_pointers_directly_in_arena_ branch from 2df9f19 to bc3f364 Compare April 16, 2026 01:15

overlookmotel marked this pull request as ready for review April 16, 2026 01:18

overlookmotel requested a review from Copilot April 16, 2026 01:18

Copilot started reviewing on behalf of overlookmotel April 16, 2026 01:19 View session

Copilot AI reviewed Apr 16, 2026

View reviewed changes

Comment thread crates/oxc_allocator/src/arena/alloc_impl.rs Outdated

Comment thread crates/oxc_allocator/src/arena/from_raw_parts.rs Outdated

Comment thread crates/oxc_allocator/src/arena/mod.rs

Comment thread crates/oxc_allocator/src/arena/mod.rs

overlookmotel marked this pull request as draft April 16, 2026 02:16

overlookmotel force-pushed the om/04-15-perf_allocator_store_pointers_directly_in_arena_ branch from bc3f364 to 421a549 Compare April 16, 2026 13:51

overlookmotel requested a review from Copilot April 16, 2026 13:53

Copilot started reviewing on behalf of overlookmotel April 16, 2026 13:53 View session

Copilot AI reviewed Apr 16, 2026

View reviewed changes

overlookmotel marked this pull request as ready for review April 16, 2026 14:03

overlookmotel added the 0-merge Merge with Graphite Merge Queue label Apr 16, 2026

graphite-app Bot force-pushed the om/04-15-perf_allocator_store_pointers_directly_in_arena_ branch from 421a549 to be2b392 Compare April 16, 2026 14:10

graphite-app Bot merged commit be2b392 into main Apr 16, 2026
27 checks passed

graphite-app Bot deleted the om/04-15-perf_allocator_store_pointers_directly_in_arena_ branch April 16, 2026 14:15

graphite-app Bot removed the 0-merge Merge with Graphite Merge Queue label Apr 16, 2026

Dunqing mentioned this pull request Apr 17, 2026

fix(isolated_declarations): drop required type check for private parameter properties on private constructors #21515

Merged

2 tasks

Dunqing mentioned this pull request Apr 17, 2026

fix(isolated_declarations): collect types from private accessors for paired inference #21516

Merged

2 tasks

oxc-guard Bot mentioned this pull request Apr 20, 2026

release(crates): oxc v0.127.0 #21572

Merged

This was referenced Apr 23, 2026

Revamp allocator #20513

Open

refactor(allocator): remove sentinel empty chunk #21775

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

perf(allocator): store pointers directly in `Arena`#21483

perf(allocator): store pointers directly in `Arena`#21483
graphite-app[bot] merged 1 commit intomainfrom
om/04-15-perf_allocator_store_pointers_directly_in_arena_

overlookmotel commented Apr 15, 2026 •

edited

Loading

Uh oh!

overlookmotel commented Apr 15, 2026

Uh oh!

codspeed-hq Bot commented Apr 15, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

overlookmotel commented Apr 16, 2026 •

edited by graphite-app Bot

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

overlookmotel commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

overlookmotel commented Apr 15, 2026

How to use the Graphite Merge Queue

Uh oh!

codspeed-hq Bot commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Footnotes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

overlookmotel commented Apr 16, 2026 • edited by graphite-app Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merge activity

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

overlookmotel commented Apr 15, 2026 •

edited

Loading

codspeed-hq Bot commented Apr 15, 2026 •

edited

Loading

overlookmotel commented Apr 16, 2026 •

edited by graphite-app Bot

Loading