Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
@ftynse I think we could simplify some genuflexions here around wrapping payload in a transform.target_tag="payload" if we had a transform.module op.
Thoughts ?
It would make it impossible to target a single op, rather than a region attached to the module op. It could also interfere with verification rules in some cases (e.g. when we cannot just inject random module around an op, such as GPU module/kernel verification). I really don't see a problem with attaching an attribute for debugging purposes.
mlir/test/Dialect/LLVM/lower-to-llvm-e2e.mlir | ||
---|---|---|
57 |
|
@ftynse I think we could simplify some genuflexions here around wrapping payload in a transform.target_tag="payload" if we had a transform.module op.
Thoughts ?