[Mlir-commits] [mlir] [mlir] Add direct vectorization lowering for `tensor.pack` ops (PR #78660)

[llvmlistbot at llvm.org]

Max191 wrote:

> What is the plan for this then? IIRC we needed direct vectorization here not only for performance but also because the vector code for this op was canonicalized away and we ended up with scalar data copies. It would be great to have some clarity as @bviyer also has the unpack counterpart here: #76087

I'm going to pick this back up tomorrow and address comments so we can land it soon.

https://github.com/llvm/llvm-project/pull/78660