[Mlir-commits] [mlir] [mlir][linalg] Implement TilingInterface for winograd operators (PR #96184)

[llvmlistbot at llvm.org]

================
@@ -8,24 +8,43 @@ func.func @tile_winograd_filter(%arg0: tensor<2x3x3x5xf32>, %arg1: tensor<6x6x5x
 module attributes {transform.with_named_sequence} {
   transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
     %0 = transform.structured.match ops{["linalg.winograd_filter_transform"]} in %arg1 : (!transform.any_op) -> !transform.any_op
-    %1, %loop1:2 = transform.structured.tile_using_for %0 tile_sizes [1, 1] : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op)
+    %1, %loop1:4 = transform.structured.tile_using_for %0 tile_sizes [1, 1, 1, 1] : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op)
----------------
Max191 wrote:

Same with the tests in this file. Can you update the outer tile sizes to be zero for the alpha dimensions?

https://github.com/llvm/llvm-project/pull/96184