You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
[CI] Increases the number of lit workers based on the GPU arch. (#1919)
* [CI] Increases the number of lit workers based on the GPU arch.
- For MFMA architectures (gfx908, gfx90a, gfx942, gfx950),
the worker count is increased to 64.
- For Navi architectures, the worker count is increased to 16.
The goal is to evaluate how CI machines handle more threads,
the tests executions remain stable under these settings.
This change is an initial step to experiment with performance
and stability improvements during CI runs.
* Limit LIT worker count for unstable CI architectures.
- Reduce LIT worker count to 20 for gfx90a/gfx908 due to CI
instability and test timeouts.
- Set worker count to 8 for Navi3x/Navi4x due to instability
on some navi CI machines.
* Temporarily disabled data-after-header validation from tuning
file check.
0 commit comments