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

[Stella Laurenzo]

================
@@ -2734,6 +2734,113 @@ FailureOr<SmallVector<Value>> SoftmaxOp::decomposeOperation(OpBuilder &b) {
   return SmallVector<Value>{result};
 }
 
+//===----------------------------------------------------------------------===//
----------------
stellaraccident wrote:

It has a tendency to not get done and not having the shape handling can mask other issues. Usually it's better to do it right, do it once.

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