[Mlir-commits] [mlir] [mlir][linalg] Vectorize directly to a named contraction (PR #147296)

[Adam Siemieniuk]

adam-smnk wrote:

For the next steps, I'd love to hear your thoughts on how broadcasts should be handled.
About making it the default, I feel it might be better to give a PSA and a bit of time before flipping the switch while this path can be adopted and further tested by downstream users.

> we should look into removing `vector.multi_reduction`. I am not sure about that myself _just yet_

Me neither🙂
I agree we shouldn't have to go through reduction and raising in presence of clear contractions. But I imagine the mutli reduction is still useful for representing, well... Just reductions.


https://github.com/llvm/llvm-project/pull/147296