[Mlir-commits] [mlir] [mlir][linalg] Enable scalable vectorization of linalg.unpack (PR #149293)

[Andrzej Warzyński]

banach-space wrote:

> I was thinking the same issue, i.e., require input vector size for both read and write, when I worked on pad op vectorization. I was thinking to provide two vector list, but appending one to the other one looks okay to me. Because it is op's implementation details, like tile sizes in tiling implementation.
> 
> Thanks for the patch, and I'll spend some time to integrate this into IREE: https://github.com/iree-org/iree/blob/a260a5e4c3033ed2aa35498865b856e68340b7dc/compiler/src/iree/compiler/Codegen/Utils/Utils.cpp#L1784C15-L1822

Thanks, let me know if you need any help 🙏🏻 

> 
> > This patch addresses the issue by requiring explicit vector sizes for both the read and write sides, enabling scalable vectorization in such cases.
> 
> We should document it somewhere. I don't find a good place, but maybe in `vectorizeUnPackOpPrecondition`'s function comment? I think it is better than nothing.

Done 👍🏻 

**UPDATE 28/7/25**
Rebased on top of main and incorporated all the required changes. All comments should be addressed. Thanks for the reviews!

https://github.com/llvm/llvm-project/pull/149293