[Mlir-commits] [mlir] [mlir][Vector] Reject scalable outer dim in multi_reduction unrolling (PR #224174)

[Cheng-Chieh Liu]

jay87406 wrote:

@FedericoBruzzone yeah that makes sense, a single scalable dim does seem doable here — could probably just run `vector.reduction`/`elementwise` ops on the whole scalable vector instead of unrolling per lane, same as the other strategy already does. Let's see what @banach-space  thinks first, can add a `TODO` or take a stab at it once there's a direction.

https://github.com/llvm/llvm-project/pull/224174