[Mlir-commits] [mlir] [mlir][vector] Reject a 0-d operand in the outerproduct parser (PR #219917)

[Diego Caballero]

dcaballe wrote:

Interesting... `vector.contract` masks are quite an old feature that was removed years ago... :)

Regarding dropping 0-d support from this op, I think we didn't reach an agreement on how to go about 0-d vectors in the Vector dialect, did we? I'm reluctant to move forward with this feature drop that might catch some people by surprise (and break their projects), even though I support this direction.

Perhaps a less controversial change here would be to just infer dim size 1 from 0-d vectors when computing the result type?

https://github.com/llvm/llvm-project/pull/219917