[Mlir-commits] [mlir] [MLIR] Refactor to create vectorization convOp precondition check (PR #130181)

[Andrzej Warzyński]

banach-space wrote:

Thanks!

> For a 2d conv op like linalg.conv_2d_nchw_fchw, this will return failure because both lhs and res shape rank is 4, not 1 or 3. Do you think this is good enough condition?

The condition looks good, but an additional comment would be helpful. It's not immediately obvious that that logic will reject 2D and 3D variants.

> Please feel free to leave additional feedbacks if there's any. If not, I plan to land this around next Monday.

SG, I won't be adding any more comments beyond the one above.

Thanks for taking care of this 🙏🏻

https://github.com/llvm/llvm-project/pull/130181