Common LAVA interface for GPU perf
For supporting method to display GPU perf events to LAVA dashboard, it needs to provide common interface for any GPUs, which started with Mali at the first step as a pilot project.
Blueprint information
- Status:
- Started
- Approver:
- Jesse Barker
- Priority:
- Undefined
- Drafter:
- Chunsang Jeong
- Direction:
- Needs approval
- Assignee:
- Chunsang Jeong
- Definition:
- New
- Series goal:
- Accepted for trunk
- Implementation:
-
Started
- Milestone target:
-
backlog
- Started by
- Chunsang Jeong
- Completed by
Related branches
Related bugs
Sprints
Whiteboard
Headline: Displying GPU perf result on LAVA dashboard by using out-of-tree
Acceptance: Finished setting up server-
[chunsang 19Jan2012]: Be informed by ARM that .py weren't allowed to be opened to public.
[chunsang 24Jan2012]: Continued "Create test definition" from 2012.01 milestone.
[chunsang 03Feb2012]: Created a LAVA for Android guide at;
https:/
[chunsang 27Feb2012]: Starting on Android Origen. (It's only available platform with Mali accelerated now)
e
[chunsang 17Apr2012]: Changed perf events to be enabled at any moment by adding activate/deactivate getting perf events from driver with debugfs. Considering if it's possible to show result graphically.
Work Items
Work items:
Create test definition: DONE
Screening out viable perf data for providing common interface: DONE
Make sure .py from ARM can be integrated into LAVA: DONE
Set up server-
Add calculation source to Mali kernel driver for getting profiling data: DONE
Add workqueue (or timer) to print out the profiling result for LAVA: DONE
Add memory usage source to Mali kernel driver: DONE
Dependency tree
* Blueprints in grey have been implemented.
