[mlir][Vector] Make vector.contract work with scalable vectors

[banach-space]

This is just a small fix that makes sure that vector.contract works with scalable vectors.

Rather than duplicating all the roundtrip tests for vector.contract, I'm treating scalable vectors as an edge case and just adding a couple to verify that this works.

[c-rhodes]

LGTM at a glance but I wonder if it's worth adding integration tests? Perhaps copying mlir/test/Integration/Dialect/Vector/CPU/test-contraction.mlir for mlir/test/Integration/Dialect/Vector/CPU/ArmSVE/test-contraction.mlir?

[banach-space]

Thanks for taking a look!

LGTM at a glance but I wonder if it's worth adding integration tests? Perhaps copying mlir/test/Integration/Dialect/Vector/CPU/test-contraction.mlir for mlir/test/Integration/Dialect/Vector/CPU/ArmSVE/test-contraction.mlir?

Big +1 :)

However, I'm discovering quite a few subtle bits like this one while integrating things in IREE. Mindful that vector.contract is rather complex, I prefer doing this incrementally. But yes, we should be adding integration tests for vector.contract for SVE ASAP.

[c-rhodes]

Thanks for taking a look!

LGTM at a glance but I wonder if it's worth adding integration tests? Perhaps copying mlir/test/Integration/Dialect/Vector/CPU/test-contraction.mlir for mlir/test/Integration/Dialect/Vector/CPU/ArmSVE/test-contraction.mlir?

Big +1 :)

However, I'm discovering quite a few subtle bits like this one while integrating things in IREE. Mindful that vector.contract is rather complex, I prefer doing this incrementally. But yes, we should be adding integration tests for vector.contract for SVE ASAP.

Ok that's fine but please update the title of the commit / pull request to clarify this is simply fixing scalable dims being dropped in vector.contract lowering, because it currently implies the scope is grander than it is, and like you say this is a complex op so I doubt scalable vector support ends here?