[Mlir-commits] [mlir] [NFC][MLIR] Document better linalg morphism (PR #154313)

[Rolf Morel]

rolfmorel wrote:

I will just point out that in the context of just linalg's _structured_ ops, the term "category" could/should be "einsum" - that's the kind of operation modelled by `linalg.contract` and `linalg.elementwise` together. 

The term "category" seems more helpful in the context of there being a "category" (i.e. a tree) of structured ops and there being a distinct "category" (i.e. tree, though potentially a trivial one) for convolution ops.

Wouldn't want to be missing the forest due to thinking we are seeing just a single tree 😉 

https://github.com/llvm/llvm-project/pull/154313