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

[Andrzej Warzyński]

banach-space wrote:

> @banach-space @dcaballe @hanhanW Any preferences on how to proceed here?

The approach that you are taking here so far makes sense to me. The next step would be to:
* Enable this "specialised" path in our e2e tests. This might be mostly for SVE + SME ATM 🤔 

I can try running these and see what happens (independently of this PR). I would also like to have a quick look at the perf impact, if any. For that I would use IREE. Again, I would focus on SVE + SME. @hanhanW , should we do more testing? If yes, would you be able to help?

Thanks again for pushing on this, @adam-smnk ! 🙏🏻 

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