[Mlir-commits] [mlir] [MLIR][Linalg] Add aggregate ops decomposition pass and softmax decom… (PR #97582)

[Renato Golin]

rengolin wrote:

> 2\. If named ops didnt have (what I consider) an outdated semantics for broadcast handling, then there would be a path where you could just lower to named ops without structurally changing the lowering. That is also another path to make this work. This is also worth doing, but requires a broader agreement on direction.

This is the key here: define the semantics of the named ops. Landing the PR is secondary.

Can we agree that, IFF we have broadcast/transpose semantics to the named ops, we _should_ decompose softmax to those instead of generics?

**How** we get there is a matter for another [RFC](https://discourse.llvm.org/t/rfc-transpose-attribute-for-linalg-matmul-operations/80092/), but I want to make sure our efforts there will lead to this decision agreed on a consensus.

https://github.com/llvm/llvm-project/pull/97582