Search Issue Tracker
Won't Fix
Votes
1
Found in
2020.3.39f1
2021.3.11f1
2022.1.17f1
2022.2.0b9
2023.1.0a12
Issue ID
UUM-15748
Regression
Yes
MeshSkinning.Update takes too much time to execute
How to reproduce:
1. Open the attached “2021GPUSkinning” project
2. Select “Window” → “Analysis” → “Profiler” in the main menu
3. Make sure “Record profiling information” (circle-shaped button) is enabled
4. Enter Play Mode
5. Exit Play Mode
6. Select any frame
7. Observe the length of the MeshSkinning.Update call
Expected result: MeshSkinning.Update executes quite fast (0.8 - 1.7ms)
Actual result: MeshSkinning.Update takes a long time to execute (1.3 - 2.2ms)
Reproducible with: 2020.2.0b2, 2020.3.39f1, 2021.3.11f1, 2022.1.17f1, 2022.2.0b9, 2023.1.0a12
Not reproducible with: 2020.2.0b1
Reproducible on: Windows 10
Note: During testing, MeshSkinning.Update took 0.8-1.0ms to execute before and 1.3-1.7ms after the regression on an Intel i9-10885H CPU
Add comment
All about bugs
View bugs we have successfully reproduced, and vote for the bugs you want to see fixed most urgently.
Latest issues
- ":focus" style will not be shown when the UI element is unparented and reparented and the "Focus()" method is called without the "Blur()" method being called
- Crash on BucketAllocator::Allocate when importing TextMeshPro assets
- Playmode dropdown creates continuous "Layout update is struggling" errors at certain aspect ratios
- Editor takes a long time to open VFX Graph 'Subgraph' asset when it is heavily referenced by multiple other VFX Graphs
- "Editor Event Handler Error" error is thrown when using Terrain Tool shortcuts
Resolution Note:
On 2019.4 the skinning is done using stream out in D3D11. In 2020.2.0b2.3079 a change was made that switches to the compute path instead. Testing showed significant performance gains on GPU. The stream out path was completely removed in 2022.1.0a15.2155. As a result of the change the markers and timings have changed a bit. Performance testing indeed shows a potential slow down on CPU. We already have some work in the pipeline that should bring significant performance improvements to this too.