Limiting it to tensor<?xindex> seems unnecessary. This changes some
test expectations that were relying on other folds.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
mlir/lib/Dialect/Shape/IR/Shape.cpp | ||
---|---|---|
108 | Why should this produce a shape.const_shape and not a normal tensor constant? With this setup it is not clear which constant operation will be chosen I think. |
Why should this produce a shape.const_shape and not a normal tensor constant? With this setup it is not clear which constant operation will be chosen I think.