Step towards removing the hardcoded behavior of this trait (from ODS to Python bindings) and to instead use common interface.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
mlir/include/mlir/Interfaces/InferTypeOpInterface.td | ||
---|---|---|
182 | Is this implying that SameOperandsAndResultType shouldn't be a trait, and should instead imply a specific implementation of InferTypeOpInterface? | |
mlir/test/Dialect/SPIRV/IR/bit-ops.mlir | ||
28–32 | typo: verificaiton Also, Can you expand this TODO? It isn't obvious what the ideal state is. | |
62 | Same here and elsewhere. |
mlir/include/mlir/Interfaces/InferTypeOpInterface.td | ||
---|---|---|
182 | Yes, and most places we can simply remove it. Now I will probably still keep it as a trait as folks are indexing of it. Most cases won't need it, and then this trait would just implement the methods required for InferTypeOpInterface. |
mlir/include/mlir/Interfaces/InferTypeOpInterface.td | ||
---|---|---|
182 | (to expand: it'll become an OpTraitList ODS side, C++ we'll still have it as a trait but the op will also have the infer type interface and most special casing can be removed) |
Is this implying that SameOperandsAndResultType shouldn't be a trait, and should instead imply a specific implementation of InferTypeOpInterface?