Minor display: contents optimizations #1736
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Deployment failed with the following error: View Documentation: https://vercel.com/docs/accounts/team-members-and-roles |
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D65336148 |
NickGerleman
added a commit
to NickGerleman/react-native
that referenced
this pull request
Nov 1, 2024
Summary: X-link: facebook/yoga#1736 `LayoutableChildren<yoga::Node>::Iterator` showed up to a surprising extent on a recent trace. Part of this was during pixel grid rounding, which does full tree traversal (we should fix that...), where the iterator is the first thing to read from the node. I ran Yoga microbenchmark with Yoga compiled with `-O2`, where we saw a regression of synthetic performance by ~10%, but it turns out this build also had ASAN and some other heavy bits enabled, so the real impact was quite lower (~6%). I was able to make some optimizations in the meantime against that, which still show some minor wins, reducing that overhead to ~4% in the properly optimized build (and a bit more before that). This is still measurable on the beefy server, and the code is a bit cleaner, so let's commit these! This change makes a few different optimizations 1. Removes redundant copies 2. Removes redundant index keeping 3. Mark which branches are likely vs unlikely 4. Shrink iterator size from 6 pointers to 3 pointers 5. Avoid usage in pixel grid rounding (so we don't need to have cache read for style) In "Huge nested layout" example | Before display: contents support | After display: contents support | After optimizations | | 9.84ms | 10.39ms | 10.23ms | Changelog: [Internal] Differential Revision: D65336148
Summary: X-link: facebook/react-native#47358 `LayoutableChildren<yoga::Node>::Iterator` showed up to a surprising extent on a recent trace. Part of this was during pixel grid rounding, which does full tree traversal (we should fix that...), where the iterator is the first thing to read from the node. I ran Yoga microbenchmark with Yoga compiled with `-O2`, where we saw a regression of synthetic performance by ~10%, but it turns out this build also had ASAN and some other heavy bits enabled, so the real impact was quite lower (~6%). I was able to make some optimizations in the meantime against that, which still show some minor wins, reducing that overhead to ~4% in the properly optimized build (and a bit more before that). This is still measurable on the beefy server, and the code is a bit cleaner, so let's commit these! This change makes a few different optimizations 1. Removes redundant copies 2. Removes redundant index keeping 3. Mark which branches are likely vs unlikely 4. Shrink iterator size from 6 pointers to 3 pointers 5. Avoid usage in pixel grid rounding (so we don't need to have cache read for style) In "Huge nested layout" example | Before display: contents support | After display: contents support | After optimizations | | 9.84ms | 10.39ms | 10.23ms | Changelog: [Internal] Differential Revision: D65336148
NickGerleman
force-pushed
the
export-D65336148
branch
from
53a20df to
dab6486
Compare
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D65336148 |
Contributor
|
This pull request has been merged in 8f69ac7. |
facebook-github-bot
pushed a commit
to facebook/litho
that referenced
this pull request
Nov 5, 2024
Summary: X-link: facebook/yoga#1736 X-link: facebook/react-native#47358 `LayoutableChildren<yoga::Node>::Iterator` showed up to a surprising extent on a recent trace. Part of this was during pixel grid rounding, which does full tree traversal (we should fix that...), where the iterator is the first thing to read from the node. I ran Yoga microbenchmark with Yoga compiled with `-O2`, where we saw a regression of synthetic performance by ~10%, but it turns out this build also had ASAN and some other heavy bits enabled, so the real impact was quite lower (~6%). I was able to make some optimizations in the meantime against that, which still show some minor wins, reducing that overhead to ~4% in the properly optimized build (and a bit more before that). This is still measurable on the beefy server, and the code is a bit cleaner, so let's commit these! Note that, in real scenarios, measure functions may dominate layout time, so display: contents does not mean end-to-end 4% regression, even after this change. This change makes a few different optimizations 1. Removes redundant copies 2. Removes redundant index keeping 3. Mark which branches are likely vs unlikely 4. Shrink iterator size from 6 pointers to 3 pointers 5. Avoid usage in pixel grid rounding (so we don't need to have cache read for style) In "Huge nested layout" example | Before display: contents support | After display: contents support | After optimizations | | 9.77ms | 10.39ms | 10.17ms | Changelog: [Internal] Reviewed By: rozele Differential Revision: D65336148 fbshipit-source-id: 01c592771ed7accf2d87dddd5a3a9e0225098b56
facebook-github-bot
pushed a commit
to facebook/react-native
that referenced
this pull request
Nov 5, 2024
Summary: X-link: facebook/yoga#1736 Pull Request resolved: #47358 `LayoutableChildren<yoga::Node>::Iterator` showed up to a surprising extent on a recent trace. Part of this was during pixel grid rounding, which does full tree traversal (we should fix that...), where the iterator is the first thing to read from the node. I ran Yoga microbenchmark with Yoga compiled with `-O2`, where we saw a regression of synthetic performance by ~10%, but it turns out this build also had ASAN and some other heavy bits enabled, so the real impact was quite lower (~6%). I was able to make some optimizations in the meantime against that, which still show some minor wins, reducing that overhead to ~4% in the properly optimized build (and a bit more before that). This is still measurable on the beefy server, and the code is a bit cleaner, so let's commit these! Note that, in real scenarios, measure functions may dominate layout time, so display: contents does not mean end-to-end 4% regression, even after this change. This change makes a few different optimizations 1. Removes redundant copies 2. Removes redundant index keeping 3. Mark which branches are likely vs unlikely 4. Shrink iterator size from 6 pointers to 3 pointers 5. Avoid usage in pixel grid rounding (so we don't need to have cache read for style) In "Huge nested layout" example | Before display: contents support | After display: contents support | After optimizations | | 9.77ms | 10.39ms | 10.17ms | Changelog: [Internal] Reviewed By: rozele Differential Revision: D65336148 fbshipit-source-id: 01c592771ed7accf2d87dddd5a3a9e0225098b56
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
LayoutableChildren<yoga::Node>::Iteratorshowed up to a surprising extent on a recent trace. Part of this was during pixel grid rounding, which does full tree traversal (we should fix that...), where the iterator is the first thing to read from the node.I ran Yoga microbenchmark with Yoga compiled with
-O2, where we saw a regression of synthetic performance by ~10%, but it turns out this build also had ASAN and some other heavy bits enabled, so the real impact was quite lower (~6%).I was able to make some optimizations in the meantime against that, which still show some minor wins, reducing that overhead to ~4% in the properly optimized build (and a bit more before that). This is still measurable on the beefy server, and the code is a bit cleaner, so let's commit these!
This change makes a few different optimizations
In "Huge nested layout" example
| Before display: contents support | After display: contents support | After optimizations |
| 9.84ms | 10.39ms | 10.23ms |
Changelog: [Internal]
Differential Revision: D65336148