[Mlir-commits] [mlir] [mlir][Linalg] Prevent vectorization of generic Conv with dynamic dims (PR #185415)

[Andrzej Warzyński]

================
@@ -112,6 +112,33 @@ module attributes {transform.with_named_sequence} {
 
 // -----
 
+func.func @generic_conv1d_ncw_fcw_dyn_spatial(%input: tensor<1x8x?xf16>, %filter: tensor<4x8x1xf16>, %output: tensor<1x4x?xf16>) -> tensor<1x4x?xf16> {
----------------
banach-space wrote:

[nit] Add a comment explaining _why_ there's a test using `linalg.generic`. 

https://github.com/llvm/llvm-project/pull/185415