Search Issue Tracker
Fixed in 5.6.0
Votes
7
Found in
5.4.1f1
Issue ID
843587
Regression
Yes
MeshSkinning.Update is Slowed Down by MeshSkinning.CalcMatrices
MeshSkinning.Update is Slowed Down by MeshSkinning.CalcMatrices. CPU time used by CalcMatrices is ~150% more than it used to be. (Compared to 5.3.6f1)
Reproducible on:
5.4.2f2
5.4.1f1
Not Reproducible on:
5.3.6f1
Repro Steps:
1- Download the repro project provided by user (Scratch53.zip).
2- Make copies of the project for Unity 5.4 and Unity 5.3.
3- Do the following steps using both versions of Unity.
4- Open the TrainingRoom scene, then hit Play(Or make a development build and run it).
5- Use the Profiler and look at the CPU time taken by MeshSkinning.Update->MeshSkinningPrepare->MeshSkinning.CalcMatrices (See screenshots provided by tester.)
6- Compare the results you obtained from two different versions of Unity. (Tester results are ~0.18 ms for Unity 5.4 and ~0.07 ms for Unity 5.3)
Expected Result: CPU time taken by MeshSkinning.CalcMatrices is same or less than it used to be.
What Happens: CPU time taken by MeshSkinning.CalcMatrices is ~150% more than it used to be.
Comments (1)
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
- Main Camera rendering is being overwritten by a UI Camera when using multiple cameras with URP and the second camera's background type is set to "Uninitialized"
- VisualTreeAsset has significantly increased memory allocation size at runtime when the project is built with IL2CPP scripting backend
- NullReferenceException is logged when cancelling Shader Graph recovery of an already deleted Shader Graph
- White Edges appear during Camera movement in runtime when using FSR2 with Motion Vectors and Exposure enabled
- Lens Flare renders without occlusion when the light source is obstructed by solid objects when running on Android using OpenGL
Thelo
Oct 26, 2016 23:34
I reported this issue and would like to add that the mesh skinning performance regression seems even dramatically worse on a machine using a Xeon processor, with lots of CPU cores. The machine I originally reported this from has an Intel(R) Core(TM) i5-2400 CPU @ 3.10GHz (4 CPUs), ~3.1GHz.
So as a hypothesis, this might be a multithreading issue, where 5.4 introduced a new way to deal with multiple cores that is more inefficient as the machine has more CPU cores.