Search Issue Tracker
Won't Fix
Votes
0
Found in
2021.3.33f1
2022.3.15f1
2023.2.3f1
2023.3.0a17
6000.0.0b11
Issue ID
UUM-57940
Regression
No
Application crashes or stucks with “Execution of the command buffer was aborted“ logs when stopping shader execution using UAV condition on mobile devices
How to reproduce:
1. Build And Run the user’s attached “uav_bug.zip” project on a mobile device
2. Observe the quad
Expected result: quad is green
Actual result: app crashes or stucks and logs "Execution of the command buffer was aborted due to an error during execution”
Reproducible with: 2021.3.33f1, 2022.3.15f1, 2023.2.3f1, 2023.3.0a17
Reproduced on:
VLNQA00494 - iPhone 14 Pro Max, 16.3.1 iOS
VLNQA00392 - iPad (9th generation), 15.0 iOS
VLNQA00310 - iPad Pro 12.9", 13.4.1 iOS
Android (Vulkan, OpenGLES) (user’s info)
Not reproducible on:
VLNQA00358 - iPhone 12, 14.1 iOS
Note: not reproducible in the Editor and Standalone Build (macOS)
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
- "DateTime.Now" in the UK timezone does not adjust when the British Summer Time is active
- "Unsupported source texture format (7) in ComputeNextMipLevel" error is shown and the texture atlas is corrupted when the texture format "RGB Compressed ETC2 4bits" is selected
- HDRP Graphics "High Quality Line Rendering" is missing a documentation link
- HDRP Graphics "Screen Space Global Illumination" has a link to a wrong documentation
- OnParticleUpdateJobScheduled is never called when using ParticleSystem.Simulate() in FixedUpdate()
Resolution Note:
Hello, this crash seems to happen at the GPU driver level. From looking at the provided project, it seems like the UAV condition on threadId 0 is used as a flag to early out all other compute threads within other workgroups. However, the issue here is that this can potentially cause a deadlock on some devices as there's no guarantee that all groups execute simultaneously or in a particular order. This will cause a deadlock on the threadgroups that are spinning indefinitely waiting for the flag to be set.