Batch Android animated prop updates into one JNI call per surface - #58774
Open
bartlomiejbloniarz wants to merge 3 commits into
Open
bartlomiejbloniarz wants to merge 3 commits into
bartlomiejbloniarz wants to merge 3 commits into
Conversation
added 3 commits
September 30, 2026 07:34
… animates layout Summary: The shared animation backend decided per surface whether a frame's animated props go through a shadow tree commit or straight to the mounted views. As soon as one view animated a layout prop, every animated view on that surface went through the commit path for the whole animation, including views that only animate `transform` or `opacity`. Besides the extra commit work, on Android this moved those views from `updatePropsSynchronously` to regular mounts. There the synchronous mount props cache (`overrideBySynchronousMountPropsAtMountingAndroid`) replaced the incoming transform with its last synchronously written value, so the view froze while the other view's layout animation ran and jumped when it ended. The decision is now made per view. `SurfaceUpdates` holds a surface's mutations keyed by view tag, and `applySurfaceUpdates` splits them into views with layout updates, which go through `commitUpdates`, and the rest, which are applied synchronously. `AnimationMutation` and `AnimationMutations` move to `AnimationMutation.h`, still included by `AnimationBackend.h`. `AnimatedPropsRegistry::update` reads the frame's batches instead of the merged per-surface map. ## Changelog: [General] [Fixed] - A view animating only non-layout props on the shared animation backend no longer freezes while another view on the surface animates a layout prop Differential Revision: D122570616
Summary: The animation backend hands its synchronous updates to the platform one view at a time. This adds `UIManager::synchronouslyUpdateAnimatedProps` and a matching `SchedulerDelegate` call that receive a whole frame's updates at once, continuing #56771. The backend uses it when `optimizedAnimatedPropUpdates` is on. The default delegate implementation forwards to the existing per-view update, so platforms that don't implement it are unaffected. ## Changelog: [General] [Changed] - Add a batched scheduler delegate call for animated prop updates behind `optimizedAnimatedPropUpdates` Differential Revision: D122570617
Summary: With the batched delegate from the previous diff, Android applied the backend's synchronous updates with one JNI call and one `ReadableNativeMap` per view per frame. When `optimizedAnimatedPropUpdates` is on, a surface's updates now go to the UI thread in a single call: `FabricMountingManager::synchronouslyUpdateAnimatedProps` encodes them into int and double buffers, and `BatchedAnimatedPropsMountItem` decodes and applies them. This continues the encoder and mount item from #56468. Views that are not mounted yet are still queued. A view is packed only when every prop it carries has a command. Unsupported props, null resets, platform colors, malformed values and typed props send the whole view through the existing `packAnimatedProps` serializer as a map instead. The commands cover the props Reanimated applies synchronously on Android, following the parity review in #56468 (comment). Angles are converted to radians in C++, and radii and translations keep their px or percent units. Because the decoder produces `JavaOnlyArray` transforms, `TransformHelper`'s Kotlin conversion now applies a uniform scale to Z as well as X and Y, matching the native conversion when scale is combined with 3D rotation. On a low-end Android phone (4× Cortex-A53) with 12 views animating opacity, UI-thread time per frame's synchronous batch dropped from about 6.5 ms to 3.4 ms. ## Changelog: [Android] [Performance] - Batch synchronous animated prop updates when `optimizedAnimatedPropUpdates` is enabled Differential Revision: D122570618
|
@bartlomiejbloniarz has exported this pull request. If you are a Meta employee, you can view the originating Diff in D122570618. |
This branch has not been deployed
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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:
With the batched delegate from the previous diff, Android applied the backend's synchronous updates with one JNI call and one
ReadableNativeMapper view per frame. WhenoptimizedAnimatedPropUpdatesis on, a surface's updates now go to the UI thread in a single call:FabricMountingManager::synchronouslyUpdateAnimatedPropsencodes them into int and double buffers, andBatchedAnimatedPropsMountItemdecodes and applies them. This continues the encoder and mount item from #56468. Views that are not mounted yet are still queued.A view is packed only when every prop it carries has a command. Unsupported props, null resets, platform colors, malformed values and typed props send the whole view through the existing
packAnimatedPropsserializer as a map instead. The commands cover the props Reanimated applies synchronously on Android, following the parity review in #56468 (comment). Angles are converted to radians in C++, and radii and translations keep their px or percent units. Because the decoder producesJavaOnlyArraytransforms,TransformHelper's Kotlin conversion now applies a uniform scale to Z as well as X and Y, matching the native conversion when scale is combined with 3D rotation.On a low-end Android phone (4× Cortex-A53) with 12 views animating opacity, UI-thread time per frame's synchronous batch dropped from about 6.5 ms to 3.4 ms.
Changelog:
[Android] [Performance] - Batch synchronous animated prop updates when
optimizedAnimatedPropUpdatesis enabledDifferential Revision: D122570618