Skip to content

panfrost: Fix Piglit issues that may cause flakes

Alyssa Rosenzweig requested to merge alyssa/mesa:panfrost/deflake into main

We just re-enabled the piglit G52 job. I'm not sure if this is a good idea, but at any rate, it's there and needs to be flake-free. This gets us closer to that goal. From what I can tell, there are no piglit-runner level flakes in the job due to the classic sources of flakiness (race conditions, memory snafu, etc). Instead, our Piglit test flakes seem to be the result of running multiple piglit processes in parallel via piglit-runner, with faults in one process sometimes affecting concurrent rendering in another. While this isn't great from a robustness standpoint, there isn't much we can do about in Mesa, other than make sure we don't fault in Piglit. Given this theory, this series fixes a number of GPU faults triggered in piglit on Mali-G52. On a local run, this seems to decrease the piglit-runner reported flake count from 3 to 2, which is a win!

In the CI setting, the pipeline that ran for this has 7 flakes in the piglit job, compared to 12-14 on the piglit jobs of recent commits merged to main. Not super scientific but looks like conservatively a ~30% reduction in piglit-runner level flakes for the job.

Edited by Alyssa Rosenzweig

Merge request reports