buffer: optimize Buffer.copy by ronag · Pull Request #62491 · nodejs/node

ronag · 2026-03-29T08:24:57Z

This removes some of the overhead by extending the V8 api.

buffers/buffer-copy.js n=6000000 shared='false' partial='false' bytes=1024 *** 39.65 % ±1.24% ±1.66% ±2.20%
buffers/buffer-copy.js n=6000000 shared='false' partial='false' bytes=128 *** 62.40 % ±2.66% ±3.55% ±4.64%
buffers/buffer-copy.js n=6000000 shared='false' partial='false' bytes=8 *** 75.07 % ±1.82% ±2.42% ±3.15%
buffers/buffer-copy.js n=6000000 shared='false' partial='true' bytes=1024 *** 166.82 % ±2.74% ±3.69% ±4.90%
buffers/buffer-copy.js n=6000000 shared='false' partial='true' bytes=128 *** 224.72 % ±2.15% ±2.90% ±3.83%
buffers/buffer-copy.js n=6000000 shared='false' partial='true' bytes=8 *** 253.17 % ±1.71% ±2.29% ±3.00%
buffers/buffer-copy.js n=6000000 shared='true' partial='false' bytes=1024 *** 178.14 % ±13.28% ±17.89% ±23.75%
buffers/buffer-copy.js n=6000000 shared='true' partial='false' bytes=128 *** 105.29 % ±1.83% ±2.44% ±3.18%
buffers/buffer-copy.js n=6000000 shared='true' partial='false' bytes=8 *** 75.70 % ±1.72% ±2.29% ±2.98%
buffers/buffer-copy.js n=6000000 shared='true' partial='true' bytes=1024 *** 91.38 % ±10.05% ±13.54% ±17.98%
buffers/buffer-copy.js n=6000000 shared='true' partial='true' bytes=128 *** 230.69 % ±2.03% ±2.73% ±3.62%
buffers/buffer-copy.js n=6000000 shared='true' partial='true' bytes=8 *** 251.20 % ±1.84% ±2.47% ±3.24%

// PR
node$ out/Release/node benchmark/buffers/buffer-copy.js
buffers/buffer-copy.js n=6000000 shared="true" partial="true" bytes=8: 121,402,298.52590495
buffers/buffer-copy.js n=6000000 shared="false" partial="true" bytes=8: 150,598,474.5729912
buffers/buffer-copy.js n=6000000 shared="true" partial="false" bytes=8: 158,209,945.17735347
buffers/buffer-copy.js n=6000000 shared="false" partial="false" bytes=8: 146,067,437.3395647
buffers/buffer-copy.js n=6000000 shared="true" partial="true" bytes=128: 157,367,588.5504798
buffers/buffer-copy.js n=6000000 shared="false" partial="true" bytes=128: 146,945,070.7569905
buffers/buffer-copy.js n=6000000 shared="true" partial="false" bytes=128: 145,550,697.1271931
buffers/buffer-copy.js n=6000000 shared="false" partial="false" bytes=128: 137,749,082.19508398
buffers/buffer-copy.js n=6000000 shared="true" partial="true" bytes=1024: 86,024,692.29827811
buffers/buffer-copy.js n=6000000 shared="false" partial="true" bytes=1024: 82,854,617.76403005
buffers/buffer-copy.js n=6000000 shared="true" partial="false" bytes=1024: 65,355,531.58653537
buffers/buffer-copy.js n=6000000 shared="false" partial="false" bytes=1024: 57,637,071.98025907

// v25.6.1
node$ node benchmark/buffers/buffer-copy.js
buffers/buffer-copy.js n=6000000 shared="true" partial="true" bytes=8: 42,373,280.25128937
buffers/buffer-copy.js n=6000000 shared="false" partial="true" bytes=8: 44,768,824.757315286
buffers/buffer-copy.js n=6000000 shared="true" partial="false" bytes=8: 68,458,244.01386568
buffers/buffer-copy.js n=6000000 shared="false" partial="false" bytes=8: 86,251,974.36160062
buffers/buffer-copy.js n=6000000 shared="true" partial="true" bytes=128: 43,440,905.782851204
buffers/buffer-copy.js n=6000000 shared="false" partial="true" bytes=128: 46,963,775.47802432
buffers/buffer-copy.js n=6000000 shared="true" partial="false" bytes=128: 53,035,118.51192226
buffers/buffer-copy.js n=6000000 shared="false" partial="false" bytes=128: 85,299,162.71621366
buffers/buffer-copy.js n=6000000 shared="true" partial="true" bytes=1024: 33,336,852.223290235
buffers/buffer-copy.js n=6000000 shared="false" partial="true" bytes=1024: 40,153,531.31000715
buffers/buffer-copy.js n=6000000 shared="true" partial="false" bytes=1024: 16,149,801.988758216
buffers/buffer-copy.js n=6000000 shared="false" partial="false" bytes=1024: 49,143,503.67479149

nodejs-github-bot · 2026-03-29T08:25:02Z

Review requested:

@nodejs/security-wg
@nodejs/v8-update

ronag · 2026-03-29T08:43:58Z

@nodejs/performance

This removes some of the overhead by extending the V8 api. PR-URL: nodejs#62491

mcollina

lgtm

github-actions · 2026-03-29T09:23:20Z

Failed to start CI

 ⚠ Commits were pushed since the last approving review:
⚠ - buffer: optimize Buffer.copy
⚠ - Update v8-array-buffer.h
✘ Refusing to run CI on potentially unsafe PR

https://github.com/nodejs/node/actions/runs/23705961064

ronag · 2026-03-29T09:50:49Z

If you feel strongly about the check please open a pr and fix it properly across all the other methods as well. As you wrote in the issue. Messing around with prototypes is not supported.

joyeecheung · 2026-03-29T10:22:15Z

Can you upstream the V8 changes first? In general floating patches is reserved for toolchains/platforms V8 doesn't support, that adds maintenance burden to rebase the patches during V8 updates especially when V8 does any architectural changes (e.g. they are constantly changing how handles should be used internally, which can easily turns patches like this an upgrade blocker), so it's always better to merge them first in the upstream.

ronag · 2026-03-29T10:30:34Z

https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/1815

ronag · 2026-03-29T10:32:13Z

an you upstream the V8 changes first?

I don't think it's likely to land and I don't know how to upstream to V8. If that's a blocker for you then this PR is moot. I appreciate the maintenance burden and if those that have to deal with it object then I have no objection to closing this PR.

joyeecheung

I think this needs to be upstreamed first. I could catch the pointer issue from a glance. But I don't think I know enough to gauge the internal API uses are safe and along the lines of where V8 is going without review from people in the V8 team. Floating patches can both add maintenance burden to our V8 upgrades as well as V8's integration CI (they build Node.js with different configurations there e.g. with pointer compression and could get hit by a different set of problems). They are generally reserved for things that V8 defer to others (MSVC support or niche platform support - even in those cases we still try to upstream to minimize the burden, like the big endian patches), not things like features or performance optimizations.

ronag · 2026-03-29T10:46:13Z

Understood. Unfortunate. Was a significant perf boost.

This removes some of the overhead by extending the V8 api. PR-URL: nodejs#62491

ronag · 2026-03-29T12:37:45Z

Benchmark results:

buffers/buffer-copy.js n=6000000 shared='false' partial='false' bytes=1024 *** 39.65 % ±1.24% ±1.66% ±2.20%
buffers/buffer-copy.js n=6000000 shared='false' partial='false' bytes=128 *** 62.40 % ±2.66% ±3.55% ±4.64%
buffers/buffer-copy.js n=6000000 shared='false' partial='false' bytes=8 *** 75.07 % ±1.82% ±2.42% ±3.15%
buffers/buffer-copy.js n=6000000 shared='false' partial='true' bytes=1024 *** 166.82 % ±2.74% ±3.69% ±4.90%
buffers/buffer-copy.js n=6000000 shared='false' partial='true' bytes=128 *** 224.72 % ±2.15% ±2.90% ±3.83%
buffers/buffer-copy.js n=6000000 shared='false' partial='true' bytes=8 *** 253.17 % ±1.71% ±2.29% ±3.00%
buffers/buffer-copy.js n=6000000 shared='true' partial='false' bytes=1024 *** 178.14 % ±13.28% ±17.89% ±23.75%
buffers/buffer-copy.js n=6000000 shared='true' partial='false' bytes=128 *** 105.29 % ±1.83% ±2.44% ±3.18%
buffers/buffer-copy.js n=6000000 shared='true' partial='false' bytes=8 *** 75.70 % ±1.72% ±2.29% ±2.98%
buffers/buffer-copy.js n=6000000 shared='true' partial='true' bytes=1024 *** 91.38 % ±10.05% ±13.54% ±17.98%
buffers/buffer-copy.js n=6000000 shared='true' partial='true' bytes=128 *** 230.69 % ±2.03% ±2.73% ±3.62%
buffers/buffer-copy.js n=6000000 shared='true' partial='true' bytes=8 *** 251.20 % ±1.84% ±2.47% ±3.24%

codecov · 2026-03-29T13:20:47Z

Codecov Report

❌ Patch coverage is 58.33333% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.70%. Comparing base (cc96741) to head (7da38b1).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/node_buffer.cc	50.00%	5 Missing ⚠️

Additional details and impacted files

@@ Coverage Diff @@## main #62491 +/- ##
==========================================
- Coverage 91.55% 89.70% -1.85% 
==========================================
Files 351 692 +341 Lines 147653 213981 +66328 Branches 23224 41050 +17826 ==========================================
+ Hits 135179 191948 +56769 - Misses 12217 14110 +1893 - Partials 257 7923 +7666

Files with missing lines	Coverage Δ
lib/buffer.js	`99.16% <100.00%> (+12.66%)`	⬆️
src/node_buffer.cc	`68.07% <50.00%> (ø)`

... and 461 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

jasnell · 2026-03-29T13:59:28Z

I like it, but yeah, I agree with @joyeecheung ... an attempt to upstream this into v8 should be made first.

ChALkeR · 2026-03-29T14:08:20Z

@ronag

The check is done in js.

It isn't robust (as I mentioned above).

Doing checks in fast api context is difficult since as far as I'm aware you can't throw from there.

You can crash the process (again, as mentioned), which is better than reading/writing arbitrary process memory.

ronag · 2026-03-29T16:27:22Z

I've funded work to upstream this to V8. Hopefully this will be able to land in the future without blocking nits.

ronag · 2026-04-07T19:31:53Z

@joyeecheung @jasnell I funded work to get this upstreamed, see https://chromium-review.googlesource.com/c/v8/v8/+/7735151

However, in order to land it in V8 we would probably need to use Relaxed_Memmove to keep it consistent with the V8 memory model, which as document in this PR is different than the node memory model. So I don't think we can upstream this to node and keep the perf improvement for SharedArrayBuffers.

Is floating a patch still off the table?

ronag · 2026-04-07T19:34:45Z

Alternatively, if there is some way to "automagically" replace base::Relaxed_Memmove with a regular memmove. Maybe a build step?

Route the native backing of Buffer.prototype.copy (CopyImpl, the `_copy` binding) through V8's new v8::ArrayBuffer::CopyArrayBufferBytes API (added in the preceding commit) instead of materializing an ArrayBufferViewContents and doing a manual memmove. This speeds up partial copies (sourceStart > 0). All copies now go through this binding: the previous %TypedArray%.prototype.set fast-path for whole-buffer copies is dropped, since it would throw on a detached or immutable target rather than report a 0-byte no-op, and the benchmarks below show the native path is comparable for that case. When both sides are backed by a SharedArrayBuffer the relaxed-atomic overload is used, which honors the SharedArrayBuffer memory model. Mixed copies (one SharedArrayBuffer, one ArrayBuffer) use the regular overload. copy() now reports the number of bytes actually copied, as returned by V8: 0 when the target is backed by a detached or immutable ArrayBuffer. The copy is then a no-op rather than a write to read-only memory. The native binding now plumbs byte offsets and the copied-byte count through as size_t, passed across the fast/slow API boundary as doubles (exact for integer values below 2^53), so copies that cross the 4 GiB boundary are no longer truncated to 32 bits. buffer-copy.js vs node v26.3.0 (x64, 30 runs): partial=false bytes=1024: +7.91% (***) partial=false bytes=128: +0.17% partial=false bytes=8: -0.89% partial=true bytes=1024: +22.33% (***) partial=true bytes=128: +19.58% (***) partial=true bytes=8: +18.70% (***) This supersedes the prototype in nodejs#62491, which added a bespoke ArrayBufferView::FastCopy instead of using the upstream-friendly CopyArrayBufferBytes API. Adds SharedArrayBuffer and immutable-ArrayBuffer coverage to the buffer copy tests, plus a pummel regression test for copies larger than 2^32 bytes. Refs: nodejs#55422 Signed-off-by: Robert Nagy <ronagy@icloud.com> Assisted-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Route the native backing of Buffer.prototype.copy (CopyImpl, the `_copy` binding) through V8's new v8::ArrayBuffer::CopyArrayBufferBytes API (added in the preceding commit) instead of materializing an ArrayBufferViewContents and doing a manual memmove. This speeds up partial copies (sourceStart > 0). All copies now go through this binding: the previous %TypedArray%.prototype.set fast-path for whole-buffer copies is dropped, since it would throw on a detached or immutable target rather than report a 0-byte no-op, and the benchmarks below show the native path is comparable for that case. When both sides are backed by a SharedArrayBuffer the relaxed-atomic overload is used, which honors the SharedArrayBuffer memory model. Mixed copies (one SharedArrayBuffer, one ArrayBuffer) use the regular overload. copy() now reports the number of bytes actually copied, as returned by V8: 0 when the target is backed by a detached or immutable ArrayBuffer. The copy is then a no-op rather than a write to read-only memory. The native binding now plumbs byte offsets and the copied-byte count through as size_t, passed across the fast/slow API boundary as doubles (exact for integer values below 2^53), so copies that cross the 4 GiB boundary are no longer truncated to 32 bits. buffer-copy.js vs node v26.3.0 (x64, 30 runs): partial=false bytes=1024: +7.91% (***) partial=false bytes=128: +0.17% partial=false bytes=8: -0.89% partial=true bytes=1024: +22.33% (***) partial=true bytes=128: +19.58% (***) partial=true bytes=8: +18.70% (***) This supersedes the prototype in #62491, which added a bespoke ArrayBufferView::FastCopy instead of using the upstream-friendly CopyArrayBufferBytes API. Adds SharedArrayBuffer and immutable-ArrayBuffer coverage to the buffer copy tests, plus a pummel regression test for copies larger than 2^32 bytes. Refs: #55422 Signed-off-by: Robert Nagy <ronagy@icloud.com> Assisted-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> PR-URL: #63828 Reviewed-By: Yagiz Nizipli <yagiz@nizipli.com> Reviewed-By: René <contact.9a5d6388@renegade334.me.uk>

ronag requested review from mcollina and targos March 29, 2026 08:24

nodejs-github-bot added c++ Issues and PRs that require attention from people who are familiar with C++. lib / src Issues and PRs related to general changes in the lib or src directory. needs-ci PRs that need a full CI run. labels Mar 29, 2026

ronag requested a review from joyeecheung March 29, 2026 08:27

ronag force-pushed the faster-copy branch from d63b225 to b436b28Compare March 29, 2026 08:28

ronag force-pushed the faster-copy branch from b436b28 to 3fe7e1bCompare March 29, 2026 08:31

ronag force-pushed the faster-copy branch from 3fe7e1b to f3b82d0Compare March 29, 2026 08:32

ronag force-pushed the faster-copy branch from f3b82d0 to 9e36c00Compare March 29, 2026 08:33

ronag requested review from anonrig and jasnell March 29, 2026 08:44

ronag force-pushed the faster-copy branch from 9e36c00 to 603008dCompare March 29, 2026 08:46

buffer: optimize Buffer.copy
785a564
This removes some of the overhead by extending the V8 api. PR-URL: nodejs#62491

ronag force-pushed the faster-copy branch from 603008d to 785a564Compare March 29, 2026 08:49

ChALkeR reviewed Mar 29, 2026
View reviewed changes

Comment threaddeps/v8/src/api/api.cc

mcollina approved these changes Mar 29, 2026
View reviewed changes

mcollina added the request-ci Add this label to start a Jenkins CI on a PR. label Mar 29, 2026

ronag commented Mar 29, 2026
View reviewed changes

Comment threaddeps/v8/include/v8-array-buffer.h Outdated

Update v8-array-buffer.h
415afe1

github-actionsBot added request-ci-failed An error occurred while starting CI via request-ci label, and manual interventon is needed. and removed request-ci Add this label to start a Jenkins CI on a PR. labels Mar 29, 2026

ronag added the author ready PRs that have at least one approval, no pending requests for changes, and a CI started. label Mar 29, 2026

joyeecheung reviewed Mar 29, 2026
View reviewed changes

Comment threaddeps/v8/include/v8-array-buffer.h Outdated

fixup
68bf482

ronag force-pushed the faster-copy branch from 8e7f4ed to 68bf482Compare March 29, 2026 10:27

fixup
7da38b1

ronag force-pushed the faster-copy branch from 3bf7f77 to 7da38b1Compare March 29, 2026 10:42

joyeecheung requested changes Mar 29, 2026
View reviewed changes

ronag closed this Mar 29, 2026

ronag added a commit to ronag/node that referenced this pull request Mar 29, 2026
buffer: optimize Buffer.copy
0859963
This removes some of the overhead by extending the V8 api. PR-URL: nodejs#62491

ronag mentioned this pull request Jun 10, 2026
buffer: optimize Buffer.prototype.copy #63828
Closed

Uh oh!

Uh oh!

Conversation

ronag commented Mar 29, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nodejs-github-bot commented Mar 29, 2026

Uh oh!

ronag commented Mar 29, 2026

Uh oh!

Uh oh!

mcollina left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actionsBot commented Mar 29, 2026

Uh oh!

ronag commented Mar 29, 2026

Uh oh!

joyeecheung commented Mar 29, 2026

Uh oh!

Uh oh!

ronag commented Mar 29, 2026

Uh oh!

ronag commented Mar 29, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joyeecheung left a comment

Choose a reason for hiding this comment

Uh oh!

ronag commented Mar 29, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ronag commented Mar 29, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecovBot commented Mar 29, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

jasnell commented Mar 29, 2026

Uh oh!

ChALkeR commented Mar 29, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ronag commented Mar 29, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ronag commented Apr 7, 2026

Uh oh!

ronag commented Apr 7, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

ronag commented Mar 29, 2026•
edited
Loading

mcollina left a comment •
edited
Loading

ronag commented Mar 29, 2026•
edited
Loading

ronag commented Mar 29, 2026•
edited
Loading

ronag commented Mar 29, 2026•
edited
Loading

codecovBot commented Mar 29, 2026•
edited
Loading

ChALkeR commented Mar 29, 2026•
edited
Loading

ronag commented Mar 29, 2026•
edited
Loading

ronag commented Apr 7, 2026•
edited
Loading