Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[Codegen][GPU] Handle dynamic and unaligned cases in DerivedThreadCon…
…fig (#18281) This adds a default set of tile sizes for any dynamically shaped copy/linalg ops that just uses the preferred vector size based on the element type bitwidth of the linalg op. This same logic might also be worth applying in aligned cases, but this patch is opting not to change pre-existing behavior without proper benchmarking. Additionally cleans up the tiling tests.
- Loading branch information