[Mlir-commits] [mlir] [MLIR] Allowing unsupported conv2d op to fail gracefully from vectorization (PR #130181)

[Andrzej Warzyński]

================
@@ -112,6 +112,25 @@ module attributes {transform.with_named_sequence} {
 
 // -----
 
+func.func @conv2d(%3: tensor<1x64x58x58xf32>, %4:  tensor<64x64x3x3xf32>) {
----------------
banach-space wrote:

1. Could you add more examples? IIUC, none of the 2d and 3d convs would vectorize with this.
2. [nit] Big input tensors are IMHO a distraction. Why not use smaller sizes like in other examples in this file?

https://github.com/llvm/llvm-project/pull/130181