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

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

> 
> > An op might have multiple ways of decomposition that should be controllable
> 
> Agree. This is a more generic description of what I named partial generalization to reach the same end state of the current decomposition. How about we make this the first step? I can start with an rfc to collect the requirements and we can team up on the design/implementation.

Lets leave the named ops discussion aside (there is a discussion on going here https://discourse.llvm.org/t/rfc-transpose-attribute-for-linalg-matmul-operations/80092/36?u=maheshravishankar) . But for this PR, lets maybe take a break, and an RFC to allow controlling the decomposition would be great. I confess, I have no great ideas to suggest, just some vague ones. So I am at a loss to really suggest how to do that.

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