[llvm] [AMDGPU][MISCHED] GCNBalancedSchedStrategy. (PR #66634)

Jeffrey Byrnes via llvm-commits llvm-commits at lists.llvm.org
Mon Sep 18 10:01:40 PDT 2023


https://github.com/jrbyrnes commented:

I have two remaining points to discuss.

1.
We bypass the UnclusteredHighRPStage (in its `initGCNRegion`) for regions with the following condition:

```
  if ((!DAG.RegionsWithMinOcc[RegionIdx] ||
       DAG.MinOccupancy <= InitialOccupancy) &&
      !DAG.RegionsWithExcessRP[RegionIdx])
    return false;
```
My question: for the balanced scheduler, should we actually run this phase for `RegionsWithMinOcc && !RegionsWithExcessRP`?

2.
Should we run `ClusteredLowOccStage` for _any_ region so long as we have observed an occupancy drop (at least for the experimental version)?

https://github.com/llvm/llvm-project/pull/66634


More information about the llvm-commits mailing list