+1 on a tensor dialect! I believe it will become a handy toolkit for people (me) who are working on loop transformation on tensors!
tensor.undeffor creating a tensor with undefined contents but a given shape (with appropriate discussion of course;undefis tricky).
+1 on this operation.