[Mlir-commits] [mlir] [mlir] Canonicalization pattern for 'shape.shape_of' (PR #98531)

Rafael Ubal llvmlistbot at llvm.org
Sat Jul 13 08:59:22 PDT 2024


rafaelubalmw wrote:

> I was confused about the pattern you replaced yes, I think it was due to not having type inference defined. So it did a little local type inference by canonicalizing type (same ops remain).

@jpienaar Thanks for the comment, Jacques. Just to be clear, would you like me to remove pattern `ShapeOfWithTensor` as I propose, or should I revert my change and keep the pattern? If I keep it, there should be a unit test stressing it, whose structure I'm unable to envision at the moment. Any idea of what that test would look like?

https://github.com/llvm/llvm-project/pull/98531


More information about the Mlir-commits mailing list