[Mlir-commits] [mlir] [mlir][linalg] Fix for bias handling for Winograd (PR #110331)

[llvmlistbot at llvm.org]

https://github.com/Max191 requested changes to this pull request.

Ok, this makes sense to me. I think it would be good to make it more clear that this is simply trying to respect the original init value of the op. I think it's more clear to keep the init slice as an init operand of the resulting decomposed ops, and it keeps the DPS nature of the sequence. Also, there are quite a few ops being generated by this decomposition now. It would be good to combine some of them into a single op.

https://github.com/llvm/llvm-project/pull/110331