[Mlir-commits] [mlir] [mlir][linalg] Implement Conv2D using Winograd Conv2D algorithm (PR #96181)

[llvmlistbot at llvm.org]

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

Nice improvements so far! A few comments:
1. The shapes for the filter transform op and the batch_matmul don't quite make sense to me. See the comments about the filter transform.
2. When the input image tensor is padded, the padded part of the tensor needs to be filled with zeros
3. It would be great to support NCHW conv layouts. This can come as a later PR, but would be really useful to have.

https://github.com/llvm/llvm-project/pull/96181