Search Issue Tracker
By Design
By Design in 1.8.X
Votes
1
Found in [Package]
1.8.4
Issue ID
BUR-2409
Regression
No
Burst does not auto-vectorize a method when the SinCos method is used
Reproduction steps:
1. Open the attached “ASDQWE” project
2. Open the “Asset/Scenes/SampleScene.unity” Scene
3. Open the Burst Inspector (Jobs > Burst > Open Inspector…)
4. Observe the “SinCos128Floats” method
Expected result: The method is automatically vectorized
Actual result: The method is not automatically vectorized and remains unoptimized
Reproducible with: 1.7.4 (2021.3.27f1), 1.8.3(2021.3.27f1), 1.8.4 (2022.3.1f1, 2023.1.0b21, 2023.2.0a18)
Reproducible on: Intel MacOS 13.3.1
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
- Out-of-bounds memory access with multiple CanvasRenderers under a Canvas when using Mesh API
- Inspector tries to access file after it was deleted when the file was locked in Inspector window
- Changing Transform values in Search window Inspector loses focus while dragging and stopping mouse without releasing dragging action
- Saving changes on the dirty VFX Graph during the Play mode throws "The referenced script (Unknown) on this Behaviour is missing!" warnings
- VFX Graph Debug Info overlaps the "Initialize" block debug info by default
Resolution Note:
This is, unfortunately, by-design. LLVM's loop vectorizer is currently unable to vectorize calls that have pointer-type parameters, such as sincos.
There's some initial community [effort|https://reviews.llvm.org/D116879] to improve this situation, but that PR is from 2022 and hasn't landed yet. If LLVM gains the ability to vectorize such calls, then we'd be able to make use of it in Burst.
Resolution Note (1.8.X):
This is, unfortunately, by-design. LLVM's loop vectorizer is currently unable to vectorize calls that have pointer-type parameters, such as sincos.
There's some initial community [effort|https://reviews.llvm.org/D116879] to improve this situation, but that PR is from 2022 and hasn't landed yet. If LLVM gains the ability to vectorize such calls, then we'd be able to make use of it in Burst.