[Mlir-commits] [mlir] [Linalg] Fix SoftmaxOp's reify result shape calculation (PR #67790)

[Abhishek Varma]

Abhishek-Varma wrote:

> Hi @Abhishek-Varma,
> 
> Could you add a test case demonstrating how the previous code was broken?

Hi @qcolombet 

So, there is ambiguity involved in the naming of `LinalgOps`.
Although `SoftmaxOp` is part of `Linalg` - it is not a `linalg.generic`, it is an aggregate op.

Therefore the `cast` of the same on `LinalgOp` is wrong and crashes.

I've now demonstrated the same by adding a test case. :)

There is a better explanation [here](https://github.com/openxla/iree/pull/15021#discussion_r1336236244) by @MaheshRavishankar .

https://github.com/llvm/llvm-project/pull/67790