[mlir][linalg] Support for vectorization for pooling ops

[vmurali]

Summary

Added support to vectorize pooling ops, piggybacking on vectorization for convolution.

[vmurali]

Clang format

[hanchung]

Can you split it into 3 patches? I think they can be

1. Add pooling named ops changes.
2. DownscaleSizeOneWindowed2DConvolution for pooling ops changes.
3. Add vectorization support for pooling ops.

Also, are the changes about convolution indexing maps needed...?

[vmurali]

In D140182#4000260, @hanchung wrote:

Can you split it into 3 patches? I think they can be

1. Add pooling named ops changes.
2. DownscaleSizeOneWindowed2DConvolution for pooling ops changes.
3. Add vectorization support for pooling ops.

Yes.

Also, are the changes about convolution indexing maps needed...?

Those got changed when I auto generated the yaml file from the python script. My guess is someone manually edited the yaml file at some point instead of using the python script. I can revert those changes in convolution ops if you want me to.

[vmurali]

This differential can be replaced by the following 3 chained differentials (descendent first) as per @hanchung 's suggestion:

https://reviews.llvm.org/D140188
https://reviews.llvm.org/D140187
https://reviews.llvm.org/D140186

Not sure which of the two options for differentials is easier to review, so not retiring this differential yet.

mlir/include/mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yaml

 metadata: !LinalgOpMetadata
   name: batch_reduce_matmul
   cpp_class_name: BatchReduceMatmulOp
   doc: |-
     Performs a batch-reduce matrix multiplication of two 3D inputs.
     The partial multiplication results are reduced into a 2D output.
 
     Numeric casting is performed on the operands to the inner multiply, promoting
-    them to the same data type as the accumulator/output."
+    them to the same data type as the accumulator/output.
   implements:
   - LinalgContractionOpInterface
 structured_op: !LinalgStructuredOpConfig
   args:
   - !LinalgOperandDefConfig
     name: A
     kind: input_tensor
     type_var: T1
metadata: !LinalgOpMetadata
   implements:
   - LinalgConvolutionOpInterface
 structured_op: !LinalgStructuredOpConfig
   args:
   - !LinalgOperandDefConfig
     name: I
     kind: input_tensor
     type_var: T1
-    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s0, s9, s1 *
-      s2 + s3 * s4, s5 * s6 + s7 * s8)>
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s0, s1, s2
+      * s3 + s4 * s5, s6 * s7 + s8 * s9)>
   - !LinalgOperandDefConfig
     name: K
     kind: input_tensor
     type_var: T2
-    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s9, s3, s7)>
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s1, s4, s8)>
   - !LinalgOperandDefConfig
     name: O
     kind: output_tensor
     type_var: U
-    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s0, s9, s1, s5)>
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s0, s1, s2,
+      s6)>
   - !LinalgOperandDefConfig
     name: strides
     kind: index_attr
-    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s2,
-      s6)>
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s3,
+      s7)>
     default_indices:
     - 1
     - 1
   - !LinalgOperandDefConfig
     name: dilations
     kind: index_attr
-    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s4,
-      s8)>
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9] -> (s5,
+      s9)>
     default_indices:
     - 1
     - 1
   indexing_maps: !LinalgIndexingMapsConfig
     static_indexing_maps:
     - affine_map<(d0, d1, d2, d3, d4, d5)[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9]
-      -> (d0, d3, d1 * s2 + d4 * s4, d2 * s6 + d5 * s8)>
+      -> (d0, d3, d1 * s3 + d4 * s5, d2 * s7 + d5 * s9)>
     - affine_map<(d0, d1, d2, d3, d4, d5)[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9]
       -> (d3, d4, d5)>
     - affine_map<(d0, d1, d2, d3, d4, d5)[s0, s1, s2, s3, s4, s5, s6, s7, s8, s9]
       -> (d0, d3, d1, d2)>
   iterator_types:
   - parallel
   - parallel
   - parallel
value: !ScalarExpression
             kind: type
             fn_name: cast_unsigned
             type_var: U
             operands:
             - !ScalarExpression
               scalar_arg: I
 --- !LinalgOpConfig
 metadata: !LinalgOpMetadata
+  name: pooling_nwc_sum
+  cpp_class_name: PoolingNwcSumOp
+  doc: |-
+    Performs sum pooling.
+
+    Layout:
+      * Input: NWC.
+      * Kernel: W.
+
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the accumulator/output.
+  implements:
+  - LinalgConvolutionOpInterface
+structured_op: !LinalgStructuredOpConfig
+  args:
+  - !LinalgOperandDefConfig
+    name: I
+    kind: input_tensor
+    type_var: T1
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1 * s2 + s3 * s4, s5)>
+  - !LinalgOperandDefConfig
+    name: K
+    kind: input_tensor
+    type_var: T2
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s3)>
+  - !LinalgOperandDefConfig
+    name: O
+    kind: output_tensor
+    type_var: U
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s5)>
+  - !LinalgOperandDefConfig
+    name: strides
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s2)>
+    default_indices:
+    - 1
+  - !LinalgOperandDefConfig
+    name: dilations
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4)>
+    default_indices:
+    - 1
+  indexing_maps: !LinalgIndexingMapsConfig
+    static_indexing_maps:
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1 * s2 + d3 * s4,
+      d2)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d3)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2)>
+  iterator_types:
+  - parallel
+  - parallel
+  - parallel
+  - reduction
+  assignments:
+  - !ScalarAssign
+    arg: O
+    value: !ScalarExpression
+      scalar_fn:
+        kind: binary
+        fn_name: add
+        operands:
+        - !ScalarExpression
+          scalar_arg: O
+        - !ScalarExpression
+          scalar_fn:
+            kind: type
+            fn_name: cast_signed
+            type_var: U
+            operands:
+            - !ScalarExpression
+              scalar_arg: I
+--- !LinalgOpConfig
+metadata: !LinalgOpMetadata
+  name: pooling_ncw_sum
+  cpp_class_name: PoolingNcwSumOp
+  doc: |-
+    Performs sum pooling.
+
+    Layout:
+      * Input: NCW.
+      * Kernel: W.
+
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the accumulator/output.
+  implements:
+  - LinalgConvolutionOpInterface
+structured_op: !LinalgStructuredOpConfig
+  args:
+  - !LinalgOperandDefConfig
+    name: I
+    kind: input_tensor
+    type_var: T1
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s2 * s3 + s4 * s5)>
+  - !LinalgOperandDefConfig
+    name: K
+    kind: input_tensor
+    type_var: T2
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4)>
+  - !LinalgOperandDefConfig
+    name: O
+    kind: output_tensor
+    type_var: U
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s2)>
+  - !LinalgOperandDefConfig
+    name: strides
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s3)>
+    default_indices:
+    - 1
+  - !LinalgOperandDefConfig
+    name: dilations
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s5)>
+    default_indices:
+    - 1
+  indexing_maps: !LinalgIndexingMapsConfig
+    static_indexing_maps:
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2 * s3 + d3
+      * s5)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d3)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2)>
+  iterator_types:
+  - parallel
+  - parallel
+  - parallel
+  - reduction
+  assignments:
+  - !ScalarAssign
+    arg: O
+    value: !ScalarExpression
+      scalar_fn:
+        kind: binary
+        fn_name: add
+        operands:
+        - !ScalarExpression
+          scalar_arg: O
+        - !ScalarExpression
+          scalar_fn:
+            kind: type
+            fn_name: cast_signed
+            type_var: U
+            operands:
+            - !ScalarExpression
+              scalar_arg: I
+--- !LinalgOpConfig
+metadata: !LinalgOpMetadata
+  name: pooling_nwc_max
+  cpp_class_name: PoolingNwcMaxOp
+  doc: |-
+    Performs max pooling.
+
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the accumulator/output.
+  implements:
+  - LinalgConvolutionOpInterface
+structured_op: !LinalgStructuredOpConfig
+  args:
+  - !LinalgOperandDefConfig
+    name: I
+    kind: input_tensor
+    type_var: T1
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1 * s2 + s3 * s4, s5)>
+  - !LinalgOperandDefConfig
+    name: K
+    kind: input_tensor
+    type_var: T2
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s3)>
+  - !LinalgOperandDefConfig
+    name: O
+    kind: output_tensor
+    type_var: U
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s5)>
+  - !LinalgOperandDefConfig
+    name: strides
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s2)>
+    default_indices:
+    - 1
+  - !LinalgOperandDefConfig
+    name: dilations
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4)>
+    default_indices:
+    - 1
+  indexing_maps: !LinalgIndexingMapsConfig
+    static_indexing_maps:
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1 * s2 + d3 * s4,
+      d2)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d3)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2)>
+  iterator_types:
+  - parallel
+  - parallel
+  - parallel
+  - reduction
+  assignments:
+  - !ScalarAssign
+    arg: O
+    value: !ScalarExpression
+      scalar_fn:
+        kind: binary
+        fn_name: max_signed
+        operands:
+        - !ScalarExpression
+          scalar_arg: O
+        - !ScalarExpression
+          scalar_fn:
+            kind: type
+            fn_name: cast_signed
+            type_var: U
+            operands:
+            - !ScalarExpression
+              scalar_arg: I
+--- !LinalgOpConfig
+metadata: !LinalgOpMetadata
+  name: pooling_nwc_max_unsigned
+  cpp_class_name: PoolingNwcMaxUnsignedOp
+  doc: |-
+    Performs unsigned max pooling.
+
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the accumulator/output.
+  implements:
+  - LinalgConvolutionOpInterface
+structured_op: !LinalgStructuredOpConfig
+  args:
+  - !LinalgOperandDefConfig
+    name: I
+    kind: input_tensor
+    type_var: T1
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1 * s2 + s3 * s4, s5)>
+  - !LinalgOperandDefConfig
+    name: K
+    kind: input_tensor
+    type_var: T2
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s3)>
+  - !LinalgOperandDefConfig
+    name: O
+    kind: output_tensor
+    type_var: U
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s5)>
+  - !LinalgOperandDefConfig
+    name: strides
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s2)>
+    default_indices:
+    - 1
+  - !LinalgOperandDefConfig
+    name: dilations
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4)>
+    default_indices:
+    - 1
+  indexing_maps: !LinalgIndexingMapsConfig
+    static_indexing_maps:
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1 * s2 + d3 * s4,
+      d2)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d3)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2)>
+  iterator_types:
+  - parallel
+  - parallel
+  - parallel
+  - reduction
+  assignments:
+  - !ScalarAssign
+    arg: O
+    value: !ScalarExpression
+      scalar_fn:
+        kind: binary
+        fn_name: max_unsigned
+        operands:
+        - !ScalarExpression
+          scalar_arg: O
+        - !ScalarExpression
+          scalar_fn:
+            kind: type
+            fn_name: cast_unsigned
+            type_var: U
+            operands:
+            - !ScalarExpression
+              scalar_arg: I
+--- !LinalgOpConfig
+metadata: !LinalgOpMetadata
+  name: pooling_ncw_max
+  cpp_class_name: PoolingNcwMaxOp
+  doc: |-
+    Performs max pooling.
+
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the accumulator/output.
+  implements:
+  - LinalgConvolutionOpInterface
+structured_op: !LinalgStructuredOpConfig
+  args:
+  - !LinalgOperandDefConfig
+    name: I
+    kind: input_tensor
+    type_var: T1
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s2 * s3 + s4 * s5)>
+  - !LinalgOperandDefConfig
+    name: K
+    kind: input_tensor
+    type_var: T2
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4)>
+  - !LinalgOperandDefConfig
+    name: O
+    kind: output_tensor
+    type_var: U
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s2)>
+  - !LinalgOperandDefConfig
+    name: strides
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s3)>
+    default_indices:
+    - 1
+  - !LinalgOperandDefConfig
+    name: dilations
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s5)>
+    default_indices:
+    - 1
+  indexing_maps: !LinalgIndexingMapsConfig
+    static_indexing_maps:
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2 * s3 + d3
+      * s5)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d3)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2)>
+  iterator_types:
+  - parallel
+  - parallel
+  - parallel
+  - reduction
+  assignments:
+  - !ScalarAssign
+    arg: O
+    value: !ScalarExpression
+      scalar_fn:
+        kind: binary
+        fn_name: max_signed
+        operands:
+        - !ScalarExpression
+          scalar_arg: O
+        - !ScalarExpression
+          scalar_fn:
+            kind: type
+            fn_name: cast_signed
+            type_var: U
+            operands:
+            - !ScalarExpression
+              scalar_arg: I
+--- !LinalgOpConfig
+metadata: !LinalgOpMetadata
+  name: pooling_nwc_min
+  cpp_class_name: PoolingNwcMinOp
+  doc: |-
+    Performs min pooling.
+
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the accumulator/output.
+  implements:
+  - LinalgConvolutionOpInterface
+structured_op: !LinalgStructuredOpConfig
+  args:
+  - !LinalgOperandDefConfig
+    name: I
+    kind: input_tensor
+    type_var: T1
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1 * s2 + s3 * s4, s5)>
+  - !LinalgOperandDefConfig
+    name: K
+    kind: input_tensor
+    type_var: T2
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s3)>
+  - !LinalgOperandDefConfig
+    name: O
+    kind: output_tensor
+    type_var: U
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s5)>
+  - !LinalgOperandDefConfig
+    name: strides
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s2)>
+    default_indices:
+    - 1
+  - !LinalgOperandDefConfig
+    name: dilations
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4)>
+    default_indices:
+    - 1
+  indexing_maps: !LinalgIndexingMapsConfig
+    static_indexing_maps:
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1 * s2 + d3 * s4,
+      d2)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d3)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2)>
+  iterator_types:
+  - parallel
+  - parallel
+  - parallel
+  - reduction
+  assignments:
+  - !ScalarAssign
+    arg: O
+    value: !ScalarExpression
+      scalar_fn:
+        kind: binary
+        fn_name: min_signed
+        operands:
+        - !ScalarExpression
+          scalar_arg: O
+        - !ScalarExpression
+          scalar_fn:
+            kind: type
+            fn_name: cast_signed
+            type_var: U
+            operands:
+            - !ScalarExpression
+              scalar_arg: I
+--- !LinalgOpConfig
+metadata: !LinalgOpMetadata
+  name: pooling_nwc_min_unsigned
+  cpp_class_name: PoolingNwcMinUnsignedOp
+  doc: |-
+    Performs unsigned min pooling.
+
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the accumulator/output.
+  implements:
+  - LinalgConvolutionOpInterface
+structured_op: !LinalgStructuredOpConfig
+  args:
+  - !LinalgOperandDefConfig
+    name: I
+    kind: input_tensor
+    type_var: T1
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1 * s2 + s3 * s4, s5)>
+  - !LinalgOperandDefConfig
+    name: K
+    kind: input_tensor
+    type_var: T2
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s3)>
+  - !LinalgOperandDefConfig
+    name: O
+    kind: output_tensor
+    type_var: U
+    shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1, s5)>
+  - !LinalgOperandDefConfig
+    name: strides
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s2)>
+    default_indices:
+    - 1
+  - !LinalgOperandDefConfig
+    name: dilations
+    kind: index_attr
+    index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4)>
+    default_indices:
+    - 1
+  indexing_maps: !LinalgIndexingMapsConfig
+    static_indexing_maps:
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1 * s2 + d3 * s4,
+      d2)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d3)>
+    - affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3, s4, s5] -> (d0, d1, d2)>
+  iterator_types:
+  - parallel
+  - parallel
+  - parallel
+  - reduction
+  assignments:
+  - !ScalarAssign
+    arg: O
+    value: !ScalarExpression
+      scalar_fn:
+        kind: binary
+        fn_name: min_unsigned
+        operands:
+        - !ScalarExpression
+          scalar_arg: O
+        - !ScalarExpression
+          scalar_fn:
+            kind: type
+            fn_name: cast_unsigned
+            type_var: U
+            operands:
+            - !ScalarExpression
+              scalar_arg: I
+--- !LinalgOpConfig
+metadata: !LinalgOpMetadata
   name: pooling_ndhwc_sum
   cpp_class_name: PoolingNdhwcSumOp
   doc: |-
     Performs 3D sum pooling.
 
     Numeric casting is performed on the input operand, promoting it to the same
     data type as the accumulator/output.
   implements:

mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp

if (succeeded(windowedNchw)) {
     return DiagnosedSilenceableFailure::success();
   }
   FailureOr<LinalgOp> depthwise =
       tryApply<DownscaleDepthwiseConv2DNhwcHwcOp>(target);
   if (succeeded(depthwise)) {
     results.push_back(*depthwise);
     return DiagnosedSilenceableFailure::success();
   }
+  FailureOr<LinalgOp> poolingNhwcSum = tryApply<
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcSumOp, PoolingNwcSumOp>>(
+      target);
+  if (succeeded(poolingNhwcSum)) {
+    results.push_back(*poolingNhwcSum);
+    return DiagnosedSilenceableFailure::success();
+  }
+  FailureOr<LinalgOp> poolingNchwSum = tryApply<
+      DownscaleSizeOneWindowed2DConvolution<PoolingNchwSumOp, PoolingNcwSumOp>>(
+      target);
+  if (succeeded(poolingNchwSum)) {
+    results.push_back(*poolingNchwSum);
+    return DiagnosedSilenceableFailure::success();
+  }
+  FailureOr<LinalgOp> poolingNhwcMax = tryApply<
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMaxOp, PoolingNwcMaxOp>>(
+      target);
+  if (succeeded(poolingNhwcMax)) {
+    results.push_back(*poolingNhwcMax);
+    return DiagnosedSilenceableFailure::success();
+  }
+  FailureOr<LinalgOp> poolingNhwcMaxUnsigned =
+      tryApply<DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMaxUnsignedOp,
+                                                     PoolingNwcMaxUnsignedOp>>(
+          target);
+  if (succeeded(poolingNhwcMaxUnsigned)) {
+    results.push_back(*poolingNhwcMaxUnsigned);
+    return DiagnosedSilenceableFailure::success();
+  }
+  FailureOr<LinalgOp> poolingNhwcMin = tryApply<
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMinOp, PoolingNwcMinOp>>(
+      target);
+  if (succeeded(poolingNhwcMin)) {
+    results.push_back(*poolingNhwcMin);
+    return DiagnosedSilenceableFailure::success();
+  }
+  FailureOr<LinalgOp> poolingNhwcMinUnsigned =
+      tryApply<DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMinUnsignedOp,
+                                                     PoolingNwcMinUnsignedOp>>(
+          target);
+  if (succeeded(poolingNhwcMinUnsigned)) {
+    results.push_back(*poolingNhwcMinUnsigned);
+    return DiagnosedSilenceableFailure::success();
+  }
+  FailureOr<LinalgOp> poolingNchwMax = tryApply<
+      DownscaleSizeOneWindowed2DConvolution<PoolingNchwMaxOp, PoolingNcwMaxOp>>(
+      target);
+  if (succeeded(poolingNchwMax)) {
+    results.push_back(*poolingNchwMax);
+    return DiagnosedSilenceableFailure::success();
+  }
   results.assign(1, nullptr);
   return emitDefaultSilenceableFailure(target);
 }
 //===----------------------------------------------------------------------===//
 // FuseOp
 //===----------------------------------------------------------------------===//
 
 /// Apply a tiling transformation to all payload ops and store both the

mlir/lib/Dialect/Linalg/Transforms/Transforms.cpp

TypeSwitch<Operation *>(convOp)
         owIndex = 2;
       })
       .Case([&](linalg::Conv2DNchwFchwOp op) {
         khIndex = 2;
         kwIndex = 3;
         ohIndex = 2;
         owIndex = 3;
       })
+      .Case([&](linalg::PoolingNhwcSumOp op) {
+        khIndex = 0;
+        kwIndex = 1;
+        ohIndex = 1;
+        owIndex = 2;
+      })
+      .Case([&](linalg::PoolingNchwSumOp op) {
+        khIndex = 0;
+        kwIndex = 1;
+        ohIndex = 2;
+        owIndex = 3;
+      })
+      .Case([&](linalg::PoolingNhwcMaxOp op) {
+        khIndex = 0;
+        kwIndex = 1;
+        ohIndex = 1;
+        owIndex = 2;
+      })
+      .Case([&](linalg::PoolingNhwcMaxUnsignedOp op) {
+        khIndex = 0;
+        kwIndex = 1;
+        ohIndex = 1;
+        owIndex = 2;
+      })
+      .Case([&](linalg::PoolingNhwcMinOp op) {
+        khIndex = 0;
+        kwIndex = 1;
+        ohIndex = 1;
+        owIndex = 2;
+      })
+      .Case([&](linalg::PoolingNhwcMinUnsignedOp op) {
+        khIndex = 0;
+        kwIndex = 1;
+        ohIndex = 1;
+        owIndex = 2;
+      })
+      .Case([&](linalg::PoolingNchwMaxOp op) {
+        khIndex = 0;
+        kwIndex = 1;
+        ohIndex = 2;
+        owIndex = 3;
+      })
       .Default([&](Operation *op) {
-        llvm_unreachable("unexpected conv2d operation.");
+        llvm_unreachable("unexpected conv2d/pool2d operation.");
       });
 
   // Only handle the case where at least one of the window dimensions is
   // of size 1. Other cases can rely on tiling to reduce to such cases.
   int64_t khSize = kernelShape[khIndex], kwSize = kernelShape[kwIndex];
   int64_t ohSize = outputShape[ohIndex], owSize = outputShape[owIndex];
   bool removeH = (khSize == 1 && ohSize == 1);
   bool removeW = (kwSize == 1 && owSize == 1);
FailureOr<Conv1DOp> DownscaleSizeOneWindowed2DConvolution<Conv2DOp, Conv1DOp>::
 
   return conv1DOp;
 }
 
 template struct linalg::DownscaleSizeOneWindowed2DConvolution<Conv2DNhwcHwcfOp,
                                                               Conv1DNwcWcfOp>;
 template struct linalg::DownscaleSizeOneWindowed2DConvolution<Conv2DNchwFchwOp,
                                                               Conv1DNcwFcwOp>;
+template struct linalg::DownscaleSizeOneWindowed2DConvolution<PoolingNhwcSumOp,
+                                                              PoolingNwcSumOp>;
+template struct linalg::DownscaleSizeOneWindowed2DConvolution<PoolingNchwSumOp,
+                                                              PoolingNcwSumOp>;
+template struct linalg::DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMaxOp,
+                                                              PoolingNwcMaxOp>;
+template struct linalg::DownscaleSizeOneWindowed2DConvolution<
+    PoolingNhwcMaxUnsignedOp, PoolingNwcMaxUnsignedOp>;
+template struct linalg::DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMinOp,
+                                                              PoolingNwcMinOp>;
+template struct linalg::DownscaleSizeOneWindowed2DConvolution<
+    PoolingNhwcMinUnsignedOp, PoolingNwcMinUnsignedOp>;
+template struct linalg::DownscaleSizeOneWindowed2DConvolution<PoolingNchwMaxOp,
+                                                              PoolingNcwMaxOp>;
 
 FailureOr<DepthwiseConv1DNwcWcOp>
 DownscaleDepthwiseConv2DNhwcHwcOp::returningMatchAndRewrite(
     DepthwiseConv2DNhwcHwcOp convOp, PatternRewriter &rewriter) const {
   if (convOp.hasBufferSemantics())
     return failure(); // To be implemented.
 
   Value input = convOp.getInputs().front();

 void linalg::populateDecomposeConvolutionPatterns(RewritePatternSet &patterns,
                                                   PatternBenefit benefit) {
   patterns.add<DownscaleSizeOneWindowed2DConvolution<linalg::Conv2DNhwcHwcfOp,
                                                      Conv1DNwcWcfOp>,
                DownscaleSizeOneWindowed2DConvolution<linalg::Conv2DNchwFchwOp,
                                                      Conv1DNcwFcwOp>,
                DownscaleDepthwiseConv2DNhwcHwcOp>(patterns.getContext(),
                                                   benefit);
+  patterns.add<
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcSumOp, PoolingNwcSumOp>,
+      DownscaleSizeOneWindowed2DConvolution<PoolingNchwSumOp, PoolingNcwSumOp>,
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMaxOp, PoolingNwcMaxOp>,
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMaxUnsignedOp,
+                                            PoolingNwcMaxUnsignedOp>,
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMinOp, PoolingNwcMinOp>,
+      DownscaleSizeOneWindowed2DConvolution<PoolingNhwcMinUnsignedOp,
+                                            PoolingNwcMinUnsignedOp>,
+      DownscaleSizeOneWindowed2DConvolution<PoolingNchwMaxOp, PoolingNcwMaxOp>>(
+      patterns.getContext(), benefit);
 }

mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp

mlir::linalg::getCombinerOpKind(Operation *combinerOp) {
   if (!combinerOp)
     return std::nullopt;
   return llvm::TypeSwitch<Operation *, llvm::Optional<CombiningKind>>(
              combinerOp)
       .Case<arith::AddIOp, arith::AddFOp>(
           [&](auto op) { return CombiningKind::ADD; })
       .Case<arith::AndIOp>([&](auto op) { return CombiningKind::AND; })
       .Case<arith::MaxSIOp>([&](auto op) { return CombiningKind::MAXSI; })
+      .Case<arith::MaxUIOp>([&](auto op) { return CombiningKind::MAXUI; })
       .Case<arith::MaxFOp>([&](auto op) { return CombiningKind::MAXF; })
       .Case<arith::MinSIOp>([&](auto op) { return CombiningKind::MINSI; })
+      .Case<arith::MinUIOp>([&](auto op) { return CombiningKind::MINUI; })
       .Case<arith::MinFOp>([&](auto op) { return CombiningKind::MINF; })
       .Case<arith::MulIOp, arith::MulFOp>(
           [&](auto op) { return CombiningKind::MUL; })
       .Case<arith::OrIOp>([&](auto op) { return CombiningKind::OR; })
       .Case<arith::XOrIOp>([&](auto op) { return CombiningKind::XOR; })
       .Default([&](auto op) { return std::nullopt; });
 }
 
Conv1DGenerator(RewriterBase &rewriter, LinalgOp linalgOp, int strideW,
     lhsShaped = linalgOp.getDpsInputOperand(0)->get();
     rhsShaped = linalgOp.getDpsInputOperand(1)->get();
     resShaped = linalgOp.getDpsInitOperand(0)->get();
     lhsShapedType = lhsShaped.getType().dyn_cast<ShapedType>();
     rhsShapedType = rhsShaped.getType().dyn_cast<ShapedType>();
     resShapedType = resShaped.getType().dyn_cast<ShapedType>();
     if (!lhsShapedType || !rhsShapedType || !resShapedType)
       return;
-    if (lhsShapedType.getRank() != 3 ||
-        (rhsShapedType.getRank() != 2 && rhsShapedType.getRank() != 3) ||
-        resShapedType.getRank() != 3)
+    if (lhsShapedType.getRank() != 3)
       return;
 
-    // Check for reduction `add` preceded by `mul`.
     Operation *reduceOp = matchLinalgReduction(linalgOp.getDpsInitOperand(0));
     if (!reduceOp)
       return;
-    llvm::Optional<vector::CombiningKind> maybeKind;
-    maybeKind = getCombinerOpKind(reduceOp);
-    if (!maybeKind || *maybeKind != vector::CombiningKind::ADD)
-      return;
-    // Check for single `mul` predecessor. The `mul` operands must be block
-    // arguments or extension of block arguments.
-    Operation *mulOp = nullptr;
+    poolRedOp = reduceOp->getName().getIdentifier();
+
+    // If (region has 2 ops (reduction + yield) or 3 ops (extension + reduction
+    // + yield) and rhs is not used) then it is the body of a pooling
+    // If conv, check for single `mul` predecessor. The `mul` operands must be
+    // block arguments or extension of block arguments.
+    // Otherwise, check for one or zero `ext` predecessor. The `ext` operands
+    // must be block arguments or extension of block arguments.
+    Operation *feedOp = nullptr;
+    bool isOnlyContinue = true;
     for (Value operand : reduceOp->getOperands()) {
-      if (operand.isa<BlockArgument>())
+      if (operand.isa<BlockArgument>()) {
         continue;
-      if (mulOp)
+      }
+      isOnlyContinue = false;
+      if (feedOp) {
         return;
-      mulOp = operand.getDefiningOp();
-      if (!mulOp || !isa<arith::MulIOp, arith::MulFOp>(mulOp))
+      }
+      feedOp = operand.getDefiningOp();
+      if (feedOp) {
+        if (isa<arith::MulIOp, arith::MulFOp>(feedOp)) {
+          for (Value localOperand : feedOp->getOperands()) {
+            if (Operation *def = localOperand.getDefiningOp()) {
+              if (!(localOperand.isa<BlockArgument>() ||
+                    (isa<CastOpInterface>(def) &&
+                     def->getOperand(0).isa<BlockArgument>()))) {
                 return;
               }
-    if (!mulOp)
+            }
+          }
+        } else if (isa<CastOpInterface>(feedOp) &&
+                   feedOp->getOperand(0).isa<BlockArgument>()) {
+          isPool = true;
+          poolExtOp = feedOp->getName().getIdentifier();
+          isPoolExt = true;
+        } else {
           return;
-    for (Value operand : mulOp->getOperands()) {
-      if (Operation *def = operand.getDefiningOp()) {
-        if (!isa<CastOpInterface>(def))
-          return;
-        operand = def->getOperand(0);
-      }
-      if (!operand.isa<BlockArgument>())
+        }
+      }
+    }
+    if (isOnlyContinue) {
+      isPool = true;
+    }
+    llvm::Optional<vector::CombiningKind> maybeKind =
+        getCombinerOpKind(reduceOp);
+    if (!(maybeKind &&
+          (*maybeKind == vector::CombiningKind::ADD ||
+           (isPool && (*maybeKind == vector::CombiningKind::MAXF ||
+                       *maybeKind == vector::CombiningKind::MAXSI ||
+                       *maybeKind == vector::CombiningKind::MAXUI ||
+                       *maybeKind == vector::CombiningKind::MINF ||
+                       *maybeKind == vector::CombiningKind::MINSI ||
+                       *maybeKind == vector::CombiningKind::MINUI))))) {
+    if (isPool) {
+      if (!(rhsShapedType.getRank() == 1 && resShapedType.getRank() == 3))
+        return;
+    } else {
+      if (!((rhsShapedType.getRank() == 2 || rhsShapedType.getRank() == 3) &&
+            resShapedType.getRank() == 3)) {
         return;
       }
+    }
     // The op is now known to be valid.
     valid = true;
   }
 
   /// Generate a vector implementation for:
   /// ```
   ///   Op def: (     n,     w,     c,    kw,    f  )
   ///    Iters: ({Par(), Par(), Par(), Red(), Red()})
   ///   Layout: {{n, strideW * w + dilationW * kw, c}, {kw, c, f}, {n, w, f}}
   /// ```
   /// kw is always unrolled.
   /// TODO: w (resp. kw) is unrolled when the strideW ( resp. dilationW) is
   /// > 1.
   FailureOr<Operation *> conv(Conv1DOpOrder conv1DOpOrder) {
-    if (!valid)
+    if (!valid) {
+      if (isPool)
+        return rewriter.notifyMatchFailure(op, "unvectorizable 1-D pool");
+      else
         return rewriter.notifyMatchFailure(op, "unvectorizable 1-D conv");
+    }
 
     int64_t nSize, wSize, cSize, kwSize, fSize;
     SmallVector<int64_t, 3> lhsShape, rhsShape, resShape;
     switch (conv1DOpOrder) {
     case Conv1DOpOrder::Nwc:
+      if (isPool) {
+        // kernel{kw}
+        bindShapeDims(rhsShapedType, kwSize);
+        // out{n, w, c}
+        bindShapeDims(resShapedType, nSize, wSize, cSize);
+        lhsShape = {nSize,
+                    // iw = ow * sw + kw *  dw - 1
+                    //   (i.e. 16 convolved with 3 (@stride 1 dilation 1) -> 14)
+                    // Perform the proper inclusive -> exclusive -> inclusive.
+                    ((wSize - 1) * strideW + 1) +
+                        ((kwSize - 1) * dilationW + 1) - 1,
+                    cSize};
+        rhsShape = {kwSize};
+        fSize = cSize;
+        resShape = {nSize, wSize, cSize};
+      } else {
         // kernel{kw, c, f}
         bindShapeDims(rhsShapedType, kwSize, cSize, fSize);
         // out{n, w, f}
         bindShapeDims(resShapedType, nSize, wSize);
         lhsShape = {nSize,
                     // iw = ow * sw + kw *  dw - 1
                     //   (i.e. 16 convolved with 3 (@stride 1 dilation 1) -> 14)
                     // Perform the proper inclusive -> exclusive -> inclusive.
-                  ((wSize - 1) * strideW + 1) + ((kwSize - 1) * dilationW + 1) -
-                      1,
+                    ((wSize - 1) * strideW + 1) +
+                        ((kwSize - 1) * dilationW + 1) - 1,
                     cSize};
         rhsShape = {kwSize, cSize, fSize};
         resShape = {nSize, wSize, fSize};
+      }
       break;
     case Conv1DOpOrder::Ncw:
+      if (isPool) {
+        // kernel{kw}
+        bindShapeDims(rhsShapedType, kwSize);
+        // out{n, c, w}
+        bindShapeDims(resShapedType, nSize, cSize, wSize);
+        lhsShape = {nSize, cSize,
+                    // iw = ow * sw + kw *  dw - 1
+                    //   (i.e. 16 convolved with 3 (@stride 1 dilation 1) -> 14)
+                    // Perform the proper inclusive -> exclusive -> inclusive.
+                    ((wSize - 1) * strideW + 1) +
+                        ((kwSize - 1) * dilationW + 1) - 1};
+        rhsShape = {kwSize};
+        fSize = cSize;
+        resShape = {nSize, cSize, wSize};
+      } else {
         // kernel{f, c, kw}
         bindShapeDims(rhsShapedType, fSize, cSize, kwSize);
         // out{n, f, w}
         bindShapeDims(resShapedType, nSize, fSize, wSize);
         lhsShape = {nSize, cSize,
                     // iw = ow * sw + kw *  dw - 1
                     //   (i.e. 16 convolved with 3 (@stride 1 dilation 1) -> 14)
                     // Perform the proper inclusive -> exclusive -> inclusive.
-                  ((wSize - 1) * strideW + 1) + ((kwSize - 1) * dilationW + 1) -
-                      1};
+                    ((wSize - 1) * strideW + 1) +
+                        ((kwSize - 1) * dilationW + 1) - 1};
         rhsShape = {fSize, cSize, kwSize};
         resShape = {nSize, fSize, wSize};
+      }
       break;
     }
 
     vector::TransferWriteOp write;
     Value zero = rewriter.create<arith::ConstantIndexOp>(loc, 0);
 
     // w is unrolled (i.e. wSizeStep == 1) iff strideW > 1.
     // When strideW == 1, we can batch the contiguous loads and avoid
     // unrolling
     int64_t wSizeStep = strideW == 1 ? wSize : 1;
 
     Type lhsEltType = lhsShapedType.getElementType();
     Type rhsEltType = rhsShapedType.getElementType();
     Type resEltType = resShapedType.getElementType();
     auto lhsType = VectorType::get(lhsShape, lhsEltType);
     auto rhsType = VectorType::get(rhsShape, rhsEltType);
     auto resType = VectorType::get(resShape, resEltType);
     // Read lhs slice of size {w * strideW + kw * dilationW, c, f} @ [0, 0,
     // 0].
     Value lhs = rewriter.create<vector::TransferReadOp>(
         loc, lhsType, lhsShaped, ValueRange{zero, zero, zero});
     // Read rhs slice of size {kw, c, f} @ [0, 0, 0].
-    Value rhs = rewriter.create<vector::TransferReadOp>(
+    Value rhs = nullptr;
+    if (!isPool) {
+      rhs = rewriter.create<vector::TransferReadOp>(
           loc, rhsType, rhsShaped, ValueRange{zero, zero, zero});
+    }
     // Read res slice of size {n, w, f} @ [0, 0, 0].
     Value res = rewriter.create<vector::TransferReadOp>(
         loc, resType, resShaped, ValueRange{zero, zero, zero});
 
     // The base vectorization case is input: {n,w,c}, weight: {kw,c,f}, output:
     // {n,w,f}. To reuse the base pattern vectorization case, we do pre
     // transpose on input, weight, and output.
     switch (conv1DOpOrder) {
     case Conv1DOpOrder::Nwc:
       // Base case, so no transposes necessary.
       break;
     case Conv1DOpOrder::Ncw: {
       // To match base vectorization case, we pre-transpose current case.
       // ncw -> nwc
       static constexpr std::array<int64_t, 3> permLhs = {0, 2, 1};
       lhs = rewriter.create<vector::TransposeOp>(loc, lhs, permLhs);
       // fcw -> wcf
       static constexpr std::array<int64_t, 3> permRhs = {2, 1, 0};
+
+      // Do not do for pooling
+      if (!isPool)
         rhs = rewriter.create<vector::TransposeOp>(loc, rhs, permRhs);
       // nfw -> nwf
       static constexpr std::array<int64_t, 3> permRes = {0, 2, 1};
       res = rewriter.create<vector::TransposeOp>(loc, res, permRes);
       break;
     }
     }
 
     //===------------------------------------------------------------------===//
     // Begin vector-only rewrite part
     //===------------------------------------------------------------------===//
     // Unroll along kw and read slices of lhs and rhs.
     SmallVector<Value> lhsVals, rhsVals, resVals;
     // Extract lhs slice of size {n, wSizeStep, c} @ [0, sw * w + dw * kw, 0].
     for (int64_t kw = 0; kw < kwSize; ++kw) {
       for (int64_t w = 0; w < wSize; w += wSizeStep) {
         lhsVals.push_back(rewriter.create<vector::ExtractStridedSliceOp>(
             loc, lhs,
             /*offsets=*/ArrayRef<int64_t>{0, w * strideW + kw * dilationW, 0},
             /*sizes=*/ArrayRef<int64_t>{nSize, wSizeStep, cSize},
             /*strides=*/ArrayRef<int64_t>{1, 1, 1}));
       }
     }
     // Extract rhs slice of size {c, f} @ [kw].
     for (int64_t kw = 0; kw < kwSize; ++kw) {
+      if (!isPool) {
         rhsVals.push_back(rewriter.create<vector::ExtractOp>(
             loc, rhs, /*offsets=*/ArrayRef<int64_t>{kw}));
       }
+    }
     // Extract res slice: {n, wSizeStep, f} @ [0, w, 0].
     for (int64_t w = 0; w < wSize; w += wSizeStep) {
       resVals.push_back(rewriter.create<vector::ExtractStridedSliceOp>(
           loc, res,
           /*offsets=*/ArrayRef<int64_t>{0, w, 0},
           /*sizes=*/ArrayRef<int64_t>{nSize, wSizeStep, fSize},
           /*strides=*/ArrayRef<int64_t>{1, 1, 1}));
     }
 
     auto linearIndex = [&](int64_t kw, int64_t w) {
       return kw * (wSize / wSizeStep) + w;
     };
 
-    // Compute contraction: O{n, w, f} += I{n, sw * w + dw * kw, c} * F{c, f}
+    // Compute contraction: O{n, w, f} += I{n, sw * w + dw * kw, c} * F{c, f} or
+    // Perform simple arith operation for pooling
     for (int64_t kw = 0; kw < kwSize; ++kw) {
       for (int64_t w = 0; w < wSize; w += wSizeStep) {
-        resVals[w] = conv1dSliceAsContraction(
-            rewriter, loc, lhsVals[linearIndex(kw, w)], rhsVals[kw], resVals[w]);
+        if (isPool) {
+          resVals[w] = pool1dSlice(rewriter, loc, lhsVals[linearIndex(kw, w)],
+                                   resVals[w]);
+        } else {
+          resVals[w] = conv1dSliceAsContraction(rewriter, loc,
+                                                lhsVals[linearIndex(kw, w)],
+                                                rhsVals[kw], resVals[w]);
+        }
       }
     }
 
     // Write back res slice: {n, wSizeStep, f} @ [0, w, 0].
     // This does not depend on kw.
     for (int64_t w = 0; w < wSize; w += wSizeStep) {
       res = rewriter.create<vector::InsertStridedSliceOp>(
           loc, resVals[w], res,
Value conv1dSliceAsContraction(RewriterBase &rewriter, Location loc,
     AffineExpr n, w, f, c;
     bindDims(ctx, n, w, f, c);
     return rewriter.create<vector::ContractionOp>(
         loc, lhs, rhs, res,
         /*indexingMaps=*/MapList{{n, w, c}, {c, f}, {n, w, f}},
         /*iteratorTypes=*/ArrayRef<vector::IteratorType>{par, par, par, red});
   }
 
+  // Create a reduction: lhs{n, w, c} -> res{n, w, c}
+  Value pool1dSlice(RewriterBase &rewriter, Location loc, Value lhs,
+                    Value res) {
+    if (isPoolExt) {
+      lhs = rewriter.create(loc, poolExtOp, lhs, res.getType())->getResult(0);
+    }
+    return rewriter
+        .create(loc, poolRedOp, ArrayRef<Value>{lhs, res}, res.getType())
+        ->getResult(0);
+  }
+
   /// Generate a vector implementation for:
   /// ```
   ///   Op def: (     n,     w,     c,    kw)
   ///    Iters: ({Par(), Par(), Par(), Red()})
   ///   Layout: {{n, strideW * w + dilationW * kw, c}, {kw, c}, {n, w, c}}
   /// ```
   /// kw is always unrolled.
   /// TODO: w (resp. kw) is unrolled when the strideW ( resp. dilationW) is
if (!iters({Par(), Par(), Par(), Red(), Red()}))
       return rewriter.notifyMatchFailure(
           op, "failed to match conv::Nwc 3-par 2-red");
 
     // No transposition needed.
     if (layout({/*lhsIndex*/ {n, strideW * w + dilationW * kw, c},
                 /*rhsIndex*/ {kw, c, f},
                 /*resIndex*/ {n, w, f}}))
       return conv(Conv1DOpOrder::Nwc);
+
     return rewriter.notifyMatchFailure(op, "not a conv::Nwc layout");
   }
 
   /// Entry point that transposes into the common form:
   ///   {{n, c, strideW * w + dilationW * kw}, {f, c, kw}, {n, f, w}}
   FailureOr<Operation *> generateNcwConv() {
     AffineExpr n, w, f, kw, c;
     bindDims(ctx, n, f, w, c, kw);
     if (!iters({Par(), Par(), Par(), Red(), Red()}))
       return rewriter.notifyMatchFailure(
           op, "failed to match conv::Ncw 3-par 2-red");
 
     if (layout({/*lhsIndex*/ {n, c, strideW * w + dilationW * kw},
                 /*rhsIndex*/ {f, c, kw},
                 /*resIndex*/ {n, f, w}}))
       return conv(Conv1DOpOrder::Ncw);
 
     return rewriter.notifyMatchFailure(op, "not a conv::Ncw layout");
   }
 
   /// Entry point that transposes into the common form:
+  ///   {{n, strideW * w + dilationW * kw, c}, {kw}, {n, w, c}} for pooling
+  FailureOr<Operation *> generateNwcPooling() {
+    AffineExpr n, w, c, kw;
+    bindDims(ctx, n, w, c, kw);
+    if (!iters({Par(), Par(), Par(), Red()}))
+      return rewriter.notifyMatchFailure(op,
+                                         "failed to match pooling 3-par 1-red");
+
+    // No transposition needed.
+    if (layout({/*lhsIndex*/ {n, strideW * w + dilationW * kw, c},
+                /*rhsIndex*/ {kw},
+                /*resIndex*/ {n, w, c}}))
+      return conv(Conv1DOpOrder::Nwc);
+
+    return rewriter.notifyMatchFailure(op, "not a pooling::Nwc layout");
+  }
+
+  /// Entry point that transposes into the common form:
+  ///   {{n, c, strideW * w + dilationW * kw}, {kw}, {n, c, w}} for pooling
+  FailureOr<Operation *> generateNcwPooling() {
+    AffineExpr n, w, c, kw;
+    bindDims(ctx, n, c, w, kw);
+    if (!iters({Par(), Par(), Par(), Red()}))
+      return rewriter.notifyMatchFailure(op,
+                                         "failed to match pooling 3-par 1-red");
+
+    if (layout({/*lhsIndex*/ {n, c, strideW * w + dilationW * kw},
+                /*rhsIndex*/ {kw},
+                /*resIndex*/ {n, c, w}}))
+      return conv(Conv1DOpOrder::Ncw);
+
+    return rewriter.notifyMatchFailure(op, "not a pooling::Ncw layout");
+  }
+
+  /// Entry point that transposes into the common form:
   ///   {{n, strideW * w + dilationW * kw, c}, {kw, c}, {n, w, c}}
   FailureOr<Operation *> generateDilatedConv() {
     AffineExpr n, w, c, kw;
     bindDims(ctx, n, w, c, kw);
     if (!iters({Par(), Par(), Par(), Red()}))
       return rewriter.notifyMatchFailure(
           op, "failed to match depthwise::Nwc conv 3-par 1-red");
 
     // No transposition needed.
     if (layout({/*lhsIndex*/ {n, strideW * w + dilationW * kw, c},
                 /*rhsIndex*/ {kw, c},
                 /*resIndex*/ {n, w, c}}))
       return depthwiseConv();
 
     return rewriter.notifyMatchFailure(op, "not a depthwise::Nwc layout");
   }
 
 private:
   bool valid = false;
+  bool isPool = false;
+  StringAttr poolRedOp;
+  StringAttr poolExtOp;
+  bool isPoolExt = false;
   int strideW, dilationW;
   Value lhsShaped, rhsShaped, resShaped;
   ShapedType lhsShapedType, rhsShapedType, resShapedType;
 };
 } // namespace
 
 /// Helper function to vectorize a LinalgOp with convolution semantics.
 // TODO: extend the generic vectorization to support windows and drop this.
static FailureOr<Operation *> vectorizeConvolution(RewriterBase &rewriter,
   auto dilation = dilations ? *dilations.getValues<uint64_t>().begin() : 1;
   Conv1DGenerator e(rewriter, op, stride, dilation);
   auto res = e.generateNwcConv();
   if (succeeded(res))
     return res;
   res = e.generateNcwConv();
   if (succeeded(res))
     return res;
+  res = e.generateNwcPooling();
+  if (succeeded(res))
+    return res;
+  res = e.generateNcwPooling();
+  if (succeeded(res))
+    return res;
   return e.generateDilatedConv();
 }
 
 struct VectorizeConvolution : public OpInterfaceRewritePattern<LinalgOp> {
   using OpInterfaceRewritePattern::OpInterfaceRewritePattern;
 
   LogicalResult matchAndRewrite(LinalgOp op,
                                 PatternRewriter &rewriter) const override {

mlir/python/mlir/dialects/linalg/opdsl/ops/core_named_ops.py

def depthwise_conv_3d_ndhwc_dhwcm(I=TensorDef(T1,
   """
   implements(ConvolutionOpInterface)
   domain(D.n, D.od, D.oh, D.ow, D.cm, D.kd, D.kh, D.kw, D.ic)
   O[D.n, D.od, D.oh, D.ow, D.ic, D.cm] += TypeFn.cast_signed(
       U, I[D.n, D.od * S.SD + D.kd * S.DD, D.oh * S.SH + D.kh * S.DH,
            D.ow * S.SW + D.kw * S.DW, D.ic]) * TypeFn.cast_signed(
                U, K[D.kd, D.kh, D.kw, D.ic, D.cm])
 
-
 @linalg_structured_op
 def pooling_nhwc_sum(I=TensorDef(T1, S.N, S.OH * S.SH + S.KH * S.DH,
                                  S.OW * S.SW + S.KW * S.DW, S.C),
                      K=TensorDef(T2, S.KH, S.KW, index_dims=[D.kh, D.kw]),
                      O=TensorDef(U, S.N, S.OH, S.OW, S.C, output=True),
                      strides=IndexAttrDef(S.SH, S.SW, default=[1, 1]),
                      dilations=IndexAttrDef(S.DH, S.DW, default=[1, 1])):
   """Performs sum pooling.
def pooling_nhwc_min_unsigned(I=TensorDef(T1, S.N, S.OH * S.SH + S.KH * S.DH,
   data type as the accumulator/output.
   """
   implements(ConvolutionOpInterface)
   domain(D.n, D.oh, D.ow, D.c, D.kh, D.kw)
   O[D.n, D.oh, D.ow,
     D.c] = ReduceFn.min_unsigned[D.kh, D.kw](TypeFn.cast_unsigned(
         U, I[D.n, D.oh * S.SH + D.kh * S.DH, D.ow * S.SW + D.kw * S.DW, D.c]))
 
+@linalg_structured_op
+def pooling_nwc_sum(I=TensorDef(T1, S.N,
+                                S.OW * S.SW + S.KW * S.DW, S.C),
+                    K=TensorDef(T2, S.KW, index_dims=[D.kw]),
+                    O=TensorDef(U, S.N, S.OW, S.C, output=True),
+                    strides=IndexAttrDef(S.SW, default=[1]),
+                    dilations=IndexAttrDef(S.DW, default=[1])):
+  """Performs sum pooling.
+
+  Layout:
+    * Input: NWC.
+    * Kernel: W.
+
+  Numeric casting is performed on the input operand, promoting it to the same
+  data type as the accumulator/output.
+  """
+  implements(ConvolutionOpInterface)
+  domain(D.n, D.ow, D.c, D.kw)
+  O[D.n, D.ow, D.c] += TypeFn.cast_signed(
+      U, I[D.n, D.ow * S.SW + D.kw * S.DW, D.c])
+
+
+@linalg_structured_op
+def pooling_ncw_sum(I=TensorDef(T1, S.N, S.C,
+                                S.OW * S.SW + S.KW * S.DW),
+                    K=TensorDef(T2, S.KW, index_dims=[D.kw]),
+                    O=TensorDef(U, S.N, S.C, S.OW, output=True),
+                    strides=IndexAttrDef(S.SW, default=[1]),
+                    dilations=IndexAttrDef(S.DW, default=[1])):
+  """Performs sum pooling.
+
+  Layout:
+    * Input: NCW.
+    * Kernel: W.
+
+  Numeric casting is performed on the input operand, promoting it to the same
+  data type as the accumulator/output.
+  """
+  implements(ConvolutionOpInterface)
+  domain(D.n, D.c, D.ow, D.kw)
+  O[D.n, D.c, D.ow] += TypeFn.cast_signed(
+      U, I[D.n, D.c, D.ow * S.SW + D.kw * S.DW])
+
+
+@linalg_structured_op
+def pooling_nwc_max(I=TensorDef(T1, S.N,
+                                S.OW * S.SW + S.KW * S.DW, S.C),
+                    K=TensorDef(T2, S.KW, index_dims=[D.kw]),
+                    O=TensorDef(U, S.N, S.OW, S.C, output=True),
+                    strides=IndexAttrDef(S.SW, default=[1]),
+                    dilations=IndexAttrDef(S.DW, default=[1])):
+  """Performs max pooling.
+
+  Numeric casting is performed on the input operand, promoting it to the same
+  data type as the accumulator/output.
+  """
+  implements(ConvolutionOpInterface)
+  domain(D.n, D.ow, D.c, D.kw)
+  O[D.n, D.ow, D.c] = ReduceFn.max_signed[[D.kw]](TypeFn.cast_signed(
+      U, I[D.n, D.ow * S.SW + D.kw * S.DW, D.c]))
+
+
+@linalg_structured_op
+def pooling_nwc_max_unsigned(I=TensorDef(T1, S.N,
+                                         S.OW * S.SW + S.KW * S.DW, S.C),
+                             K=TensorDef(T2,
+                                         S.KW,
+                                         index_dims=[D.kw]),
+                             O=TensorDef(U, S.N, S.OW, S.C, output=True),
+                             strides=IndexAttrDef(S.SW, default=[1]),
+                             dilations=IndexAttrDef(S.DW, default=[1])):
+  """Performs unsigned max pooling.
+
+  Numeric casting is performed on the input operand, promoting it to the same
+  data type as the accumulator/output.
+  """
+  implements(ConvolutionOpInterface)
+  domain(D.n, D.ow, D.c, D.kw)
+  O[D.n, D.ow,
+    D.c] = ReduceFn.max_unsigned[[D.kw]](TypeFn.cast_unsigned(
+        U, I[D.n, D.ow * S.SW + D.kw * S.DW, D.c]))
+
+
+@linalg_structured_op
+def pooling_ncw_max(I=TensorDef(T1, S.N, S.C,
+                                S.OW * S.SW + S.KW * S.DW),
+                    K=TensorDef(T2, S.KW, index_dims=[D.kw]),
+                    O=TensorDef(U, S.N, S.C, S.OW, output=True),
+                    strides=IndexAttrDef(S.SW, default=[1]),
+                    dilations=IndexAttrDef(S.DW, default=[1])):
+  """Performs max pooling.
+
+  Numeric casting is performed on the input operand, promoting it to the same
+  data type as the accumulator/output.
+  """
+  implements(ConvolutionOpInterface)
+  domain(D.n, D.c, D.ow, D.kw)
+  O[D.n, D.c, D.ow] = ReduceFn.max_signed[[D.kw]](TypeFn.cast_signed(
+      U, I[D.n, D.c, D.ow * S.SW + D.kw * S.DW,]))
+
+
+@linalg_structured_op
+def pooling_nwc_min(I=TensorDef(T1, S.N,
+                                S.OW * S.SW + S.KW * S.DW, S.C),
+                    K=TensorDef(T2, S.KW, index_dims=[D.kw]),
+                    O=TensorDef(U, S.N, S.OW, S.C, output=True),
+                    strides=IndexAttrDef(S.SW, default=[1]),
+                    dilations=IndexAttrDef(S.DW, default=[1])):
+  """Performs min pooling.
+
+  Numeric casting is performed on the input operand, promoting it to the same
+  data type as the accumulator/output.
+  """
+  implements(ConvolutionOpInterface)
+  domain(D.n, D.ow, D.c, D.kw)
+  O[D.n, D.ow, D.c] = ReduceFn.min_signed[[D.kw]](TypeFn.cast_signed(
+      U, I[D.n, D.ow * S.SW + D.kw * S.DW, D.c]))
+
+
+@linalg_structured_op
+def pooling_nwc_min_unsigned(I=TensorDef(T1, S.N,
+                                         S.OW * S.SW + S.KW * S.DW, S.C),
+                             K=TensorDef(T2,
+                                         S.KW,
+                                         index_dims=[D.kw]),
+                             O=TensorDef(U, S.N, S.OW, S.C, output=True),
+                             strides=IndexAttrDef(S.SW, default=[1]),
+                             dilations=IndexAttrDef(S.DW, default=[1])):
+  """Performs unsigned min pooling.
+
+  Numeric casting is performed on the input operand, promoting it to the same
+  data type as the accumulator/output.
+  """
+  implements(ConvolutionOpInterface)
+  domain(D.n, D.ow, D.c, D.kw)
+  O[D.n, D.ow,
+    D.c] = ReduceFn.min_unsigned[[D.kw]](TypeFn.cast_unsigned(
+        U, I[D.n, D.ow * S.SW + D.kw * S.DW, D.c]))
+
+
 
 @linalg_structured_op
 def pooling_ndhwc_sum(I=TensorDef(T1, S.N, S.OD * S.SD + S.KD * S.DD,
                                   S.OH * S.SH + S.KH * S.DH,
                                   S.OW * S.SW + S.KW * S.DW, S.C),
                       K=TensorDef(T2,
                                   S.KD,
                                   S.KH,

mlir/test/Dialect/Linalg/generalize-named-polymorphic-ops.mlir

 // CHECK-LABEL: @generalize_pooling_nhwc_max_f32
 // CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: f32, %[[SHAPE_ARG:.+]]: f32, %[[OUT_ARG:.+]]: f32)
 // CHECK-NEXT:   %[[MAX:.+]] = arith.maxf %[[OUT_ARG]], %[[IN_ARG]] : f32
 // CHECK-NEXT:   linalg.yield %[[MAX]] : f32
 // CHECK-NEXT: -> tensor<1x2x4x1xf32>
 
 // -----
 
+func.func @generalize_pooling_nwc_max_f32(%input : tensor<1x16x1xf32>, %shape: tensor<2xf32>, %output: tensor<1x4x1xf32>) -> tensor<1x4x1xf32> {
+  %0 = linalg.pooling_nwc_max {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xf32>, tensor<2xf32>) outs(%output : tensor<1x4x1xf32>) -> tensor<1x4x1xf32>
+  return %0: tensor<1x4x1xf32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_max_f32
+// CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: f32, %[[SHAPE_ARG:.+]]: f32, %[[OUT_ARG:.+]]: f32)
+// CHECK-NEXT:   %[[MAX:.+]] = arith.maxf %[[OUT_ARG]], %[[IN_ARG]] : f32
+// CHECK-NEXT:   linalg.yield %[[MAX]] : f32
+// CHECK-NEXT: -> tensor<1x4x1xf32>
+
+// -----
+
 func.func @generalize_pooling_nhwc_max_i32(%input : tensor<1x4x16x1xi32>, %shape: tensor<2x2xi32>, %output: tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32> {
   %0 = linalg.pooling_nhwc_max {dilations = dense<[1, 2]> : tensor<2xi64>, strides = dense<[2, 4]> : tensor<2xi64>}
     ins(%input, %shape : tensor<1x4x16x1xi32>, tensor<2x2xi32>) outs(%output : tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32>
   return %0: tensor<1x2x4x1xi32>
 }
 
 // CHECK-LABEL: @generalize_pooling_nhwc_max_i32
 // Verify signed integer maximum.
 // CHECK:        = arith.maxsi
 
 // -----
 
+func.func @generalize_pooling_nwc_max_i32(%input : tensor<1x16x1xi32>, %shape: tensor<2xi32>, %output: tensor<1x4x1xi32>) -> tensor<1x4x1xi32> {
+  %0 = linalg.pooling_nwc_max {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xi32>, tensor<2xi32>) outs(%output : tensor<1x4x1xi32>) -> tensor<1x4x1xi32>
+  return %0: tensor<1x4x1xi32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_max_i32
+// Verify signed integer maximum.
+// CHECK:        = arith.maxsi
+
+// -----
+
 func.func @generalize_pooling_nhwc_max_unsigned_i32(%input : tensor<1x4x16x1xi32>, %shape: tensor<2x2xi32>, %output: tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32> {
   %0 = linalg.pooling_nhwc_max_unsigned {dilations = dense<[1, 2]> : tensor<2xi64>, strides = dense<[2, 4]> : tensor<2xi64>}
     ins(%input, %shape : tensor<1x4x16x1xi32>, tensor<2x2xi32>) outs(%output : tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32>
   return %0: tensor<1x2x4x1xi32>
 }
 
 // CHECK-LABEL: @generalize_pooling_nhwc_max_unsigned_i32
 // Verify unsigned integer minimum.
 // CHECK:        = arith.maxui
 
 // -----
 
+func.func @generalize_pooling_nwc_max_unsigned_i32(%input : tensor<1x16x1xi32>, %shape: tensor<2xi32>, %output: tensor<1x4x1xi32>) -> tensor<1x4x1xi32> {
+  %0 = linalg.pooling_nwc_max_unsigned {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xi32>, tensor<2xi32>) outs(%output : tensor<1x4x1xi32>) -> tensor<1x4x1xi32>
+  return %0: tensor<1x4x1xi32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_max_unsigned_i32
+// Verify unsigned integer minimum.
+// CHECK:        = arith.maxui
+
+// -----
+
 func.func @generalize_pooling_nhwc_min_f32(%input : tensor<1x4x16x1xf32>, %shape: tensor<2x2xf32>, %output: tensor<1x2x4x1xf32>) -> tensor<1x2x4x1xf32> {
   %0 = linalg.pooling_nhwc_min {dilations = dense<[1, 2]> : tensor<2xi64>, strides = dense<[2, 4]> : tensor<2xi64>}
     ins(%input, %shape : tensor<1x4x16x1xf32>, tensor<2x2xf32>) outs(%output : tensor<1x2x4x1xf32>) -> tensor<1x2x4x1xf32>
   return %0: tensor<1x2x4x1xf32>
 }
 
 // CHECK-LABEL: @generalize_pooling_nhwc_min_f32
 // CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: f32, %[[SHAPE_ARG:.+]]: f32, %[[OUT_ARG:.+]]: f32)
 // CHECK-NEXT:   %[[MIN:.+]] = arith.minf %[[OUT_ARG]], %[[IN_ARG]] : f32
 // CHECK-NEXT:   linalg.yield %[[MIN]] : f32
 // CHECK-NEXT: -> tensor<1x2x4x1xf32>
 
 // -----
 
+func.func @generalize_pooling_nwc_min_f32(%input : tensor<1x16x1xf32>, %shape: tensor<2xf32>, %output: tensor<1x4x1xf32>) -> tensor<1x4x1xf32> {
+  %0 = linalg.pooling_nwc_min {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xf32>, tensor<2xf32>) outs(%output : tensor<1x4x1xf32>) -> tensor<1x4x1xf32>
+  return %0: tensor<1x4x1xf32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_min_f32
+// CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: f32, %[[SHAPE_ARG:.+]]: f32, %[[OUT_ARG:.+]]: f32)
+// CHECK-NEXT:   %[[MIN:.+]] = arith.minf %[[OUT_ARG]], %[[IN_ARG]] : f32
+// CHECK-NEXT:   linalg.yield %[[MIN]] : f32
+// CHECK-NEXT: -> tensor<1x4x1xf32>
+
+// -----
+
 func.func @generalize_pooling_nhwc_min_i32(%input : tensor<1x4x16x1xi32>, %shape: tensor<2x2xi32>, %output: tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32> {
   %0 = linalg.pooling_nhwc_min {dilations = dense<[1, 2]> : tensor<2xi64>, strides = dense<[2, 4]> : tensor<2xi64>}
     ins(%input, %shape : tensor<1x4x16x1xi32>, tensor<2x2xi32>) outs(%output : tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32>
   return %0: tensor<1x2x4x1xi32>
 }
 
 // CHECK-LABEL: @generalize_pooling_nhwc_min_i32
 // Verify signed integer minimum.
 // CHECK:        = arith.minsi
 
 // -----
 
+func.func @generalize_pooling_nwc_min_i32(%input : tensor<1x16x1xi32>, %shape: tensor<2xi32>, %output: tensor<1x4x1xi32>) -> tensor<1x4x1xi32> {
+  %0 = linalg.pooling_nwc_min {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xi32>, tensor<2xi32>) outs(%output : tensor<1x4x1xi32>) -> tensor<1x4x1xi32>
+  return %0: tensor<1x4x1xi32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_min_i32
+// Verify signed integer minimum.
+// CHECK:        = arith.minsi
+
+// -----
+
 func.func @generalize_pooling_nhwc_min_unsigned_i32(%input : tensor<1x4x16x1xi32>, %shape: tensor<2x2xi32>, %output: tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32> {
   %0 = linalg.pooling_nhwc_min_unsigned {dilations = dense<[1, 2]> : tensor<2xi64>, strides = dense<[2, 4]> : tensor<2xi64>}
     ins(%input, %shape : tensor<1x4x16x1xi32>, tensor<2x2xi32>) outs(%output : tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32>
   return %0: tensor<1x2x4x1xi32>
 }
 
 // CHECK-LABEL: @generalize_pooling_nhwc_min_unsigned_i32
 // Verify unsigned integer minimum.
 // CHECK:        = arith.minui
 
 // -----
 
+func.func @generalize_pooling_nwc_min_unsigned_i32(%input : tensor<1x16x1xi32>, %shape: tensor<2xi32>, %output: tensor<1x4x1xi32>) -> tensor<1x4x1xi32> {
+  %0 = linalg.pooling_nwc_min_unsigned {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xi32>, tensor<2xi32>) outs(%output : tensor<1x4x1xi32>) -> tensor<1x4x1xi32>
+  return %0: tensor<1x4x1xi32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_min_unsigned_i32
+// Verify unsigned integer minimum.
+// CHECK:        = arith.minui
+
+// -----
+
 func.func @generalize_pooling_nhwc_sum_f32(%input : tensor<1x4x16x1xf32>, %shape: tensor<2x2xf32>, %output: tensor<1x2x4x1xf32>) -> tensor<1x2x4x1xf32> {
   %0 = linalg.pooling_nhwc_sum {dilations = dense<[1, 2]> : tensor<2xi64>, strides = dense<[2, 4]> : tensor<2xi64>}
     ins(%input, %shape : tensor<1x4x16x1xf32>, tensor<2x2xf32>) outs(%output : tensor<1x2x4x1xf32>) -> tensor<1x2x4x1xf32>
   return %0: tensor<1x2x4x1xf32>
 }
 
 // CHECK-LABEL: @generalize_pooling_nhwc_sum_f32
 // CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: f32, %[[SHAPE_ARG:.+]]: f32, %[[OUT_ARG:.+]]: f32)
 // CHECK-NEXT:   %[[ADD:.+]] = arith.addf %[[OUT_ARG]], %[[IN_ARG]] : f32
 // CHECK-NEXT:   linalg.yield %[[ADD]] : f32
 // CHECK-NEXT: -> tensor<1x2x4x1xf32>
 
 // -----
 
+func.func @generalize_pooling_nwc_sum_f32(%input : tensor<1x16x1xf32>, %shape: tensor<2xf32>, %output: tensor<1x4x1xf32>) -> tensor<1x4x1xf32> {
+  %0 = linalg.pooling_nwc_sum {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xf32>, tensor<2xf32>) outs(%output : tensor<1x4x1xf32>) -> tensor<1x4x1xf32>
+  return %0: tensor<1x4x1xf32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_sum_f32
+// CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: f32, %[[SHAPE_ARG:.+]]: f32, %[[OUT_ARG:.+]]: f32)
+// CHECK-NEXT:   %[[ADD:.+]] = arith.addf %[[OUT_ARG]], %[[IN_ARG]] : f32
+// CHECK-NEXT:   linalg.yield %[[ADD]] : f32
+// CHECK-NEXT: -> tensor<1x4x1xf32>
+
+// -----
+
 func.func @generalize_pooling_nhwc_sum_i32(%input : tensor<1x4x16x1xi32>, %shape: tensor<2x2xi32>, %output: tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32> {
   %0 = linalg.pooling_nhwc_sum {dilations = dense<[1, 2]> : tensor<2xi64>, strides = dense<[2, 4]> : tensor<2xi64>}
     ins(%input, %shape : tensor<1x4x16x1xi32>, tensor<2x2xi32>) outs(%output : tensor<1x2x4x1xi32>) -> tensor<1x2x4x1xi32>
   return %0: tensor<1x2x4x1xi32>
 }
 
 // CHECK-LABEL: @generalize_pooling_nhwc_sum_i32
 // CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: i32, %[[SHAPE_ARG:.+]]: i32, %[[OUT_ARG:.+]]: i32)
 // CHECK-NEXT:   %[[ADD:.+]] = arith.addi %[[OUT_ARG]], %[[IN_ARG]] : i32
 // CHECK-NEXT:   linalg.yield %[[ADD]] : i32
 // CHECK-NEXT: -> tensor<1x2x4x1xi32>
 
 // -----
 
+func.func @generalize_pooling_nwc_sum_i32(%input : tensor<1x16x1xi32>, %shape: tensor<2xi32>, %output: tensor<1x4x1xi32>) -> tensor<1x4x1xi32> {
+  %0 = linalg.pooling_nwc_sum {dilations = dense<[2]> : tensor<1xi64>, strides = dense<[4]> : tensor<1xi64>}
+    ins(%input, %shape : tensor<1x16x1xi32>, tensor<2xi32>) outs(%output : tensor<1x4x1xi32>) -> tensor<1x4x1xi32>
+  return %0: tensor<1x4x1xi32>
+}
+
+// CHECK-LABEL: @generalize_pooling_nwc_sum_i32
+// CHECK:      ^{{.*}}(%[[IN_ARG:.+]]: i32, %[[SHAPE_ARG:.+]]: i32, %[[OUT_ARG:.+]]: i32)
+// CHECK-NEXT:   %[[ADD:.+]] = arith.addi %[[OUT_ARG]], %[[IN_ARG]] : i32
+// CHECK-NEXT:   linalg.yield %[[ADD]] : i32
+// CHECK-NEXT: -> tensor<1x4x1xi32>
+
+// -----
+
 func.func @generalize_fill_0d(%value: f64, %O: tensor<f32>) -> tensor<f32> {
   %0 = linalg.fill ins(%value: f64) outs(%O : tensor<f32>) -> tensor<f32>
   return %0: tensor<f32>
 }
 
 // CHECK-DAG: #[[$MAP0:.+]] = affine_map<() -> ()>
 
 // CHECK-LABEL: @generalize_fill_0d

mlir/test/Dialect/Linalg/named-ops.mlir

func.func @pooling_nhwc_sum_tensor(%input: tensor<1x4x4x1xf32>) -> tensor<1x2x2x1xf32> {
   %res = linalg.pooling_nhwc_sum {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x4x4x1xf32>, tensor<3x3xf32>)
     outs(%fill: tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>
   return %res : tensor<1x2x2x1xf32>
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_sum_tensor
+// CHECK:         %{{.+}} = linalg.pooling_nwc_sum
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x1xf32>, tensor<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+func.func @pooling_nwc_sum_tensor(%input: tensor<1x4x1xf32>) -> tensor<1x2x1xf32> {
+  %fake = tensor.empty() : tensor<3xf32>
+  %init = tensor.empty() : tensor<1x2x1xf32>
+  %cst = arith.constant 0.000000e+00 : f32
+  %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+  %res = linalg.pooling_nwc_sum {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x4x1xf32>, tensor<3xf32>)
+    outs(%fill: tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+  return %res : tensor<1x2x1xf32>
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_sum
 // CHECK:         linalg.pooling_nhwc_sum
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x4x1xf32>, memref<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : memref<1x2x2x1xf32>)
 func.func @pooling_nhwc_sum(%input: memref<1x4x4x1xf32>, %fake: memref<3x3xf32>, %output: memref<1x2x2x1xf32>) {
   linalg.pooling_nhwc_sum {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: memref<1x4x4x1xf32>, memref<3x3xf32>)
     outs(%output: memref<1x2x2x1xf32>)
   return
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_sum
+// CHECK:         linalg.pooling_nwc_sum
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x1xf32>, memref<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : memref<1x2x1xf32>)
+func.func @pooling_nwc_sum(%input: memref<1x4x1xf32>, %fake: memref<3xf32>, %output: memref<1x2x1xf32>) {
+  linalg.pooling_nwc_sum {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: memref<1x4x1xf32>, memref<3xf32>)
+    outs(%output: memref<1x2x1xf32>)
+  return
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nchw_sum_tensor
 // CHECK:         %{{.+}} = linalg.pooling_nchw_sum
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x1x4x4xf32>, tensor<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x1x2x2xf32>) -> tensor<1x1x2x2xf32>
 func.func @pooling_nchw_sum_tensor(%input: tensor<1x1x4x4xf32>) -> tensor<1x1x2x2xf32> {
   %fake = tensor.empty() : tensor<3x3xf32>
   %init = tensor.empty() : tensor<1x1x2x2xf32>
   %cst = arith.constant 0.000000e+00 : f32
   %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x1x2x2xf32>) -> tensor<1x1x2x2xf32>
   %res = linalg.pooling_nchw_sum {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x1x4x4xf32>, tensor<3x3xf32>)
     outs(%fill: tensor<1x1x2x2xf32>) -> tensor<1x1x2x2xf32>
   return %res : tensor<1x1x2x2xf32>
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_ncw_sum_tensor
+// CHECK:         %{{.+}} = linalg.pooling_ncw_sum
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x1x4xf32>, tensor<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x1x2xf32>) -> tensor<1x1x2xf32>
+func.func @pooling_ncw_sum_tensor(%input: tensor<1x1x4xf32>) -> tensor<1x1x2xf32> {
+  %fake = tensor.empty() : tensor<3xf32>
+  %init = tensor.empty() : tensor<1x1x2xf32>
+  %cst = arith.constant 0.000000e+00 : f32
+  %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x1x2xf32>) -> tensor<1x1x2xf32>
+  %res = linalg.pooling_ncw_sum {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x1x4xf32>, tensor<3xf32>)
+    outs(%fill: tensor<1x1x2xf32>) -> tensor<1x1x2xf32>
+  return %res : tensor<1x1x2xf32>
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nchw_sum
 // CHECK:         linalg.pooling_nchw_sum
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x1x4x4xf32>, memref<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : memref<1x1x2x2xf32>)
 func.func @pooling_nchw_sum(%input: memref<1x1x4x4xf32>, %fake: memref<3x3xf32>, %output: memref<1x1x2x2xf32>) {
   linalg.pooling_nchw_sum {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: memref<1x1x4x4xf32>, memref<3x3xf32>)
     outs(%output: memref<1x1x2x2xf32>)
   return
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_ncw_sum
+// CHECK:         linalg.pooling_ncw_sum
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x1x4xf32>, memref<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : memref<1x1x2xf32>)
+func.func @pooling_ncw_sum(%input: memref<1x1x4xf32>, %fake: memref<3xf32>, %output: memref<1x1x2xf32>) {
+  linalg.pooling_ncw_sum {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: memref<1x1x4xf32>, memref<3xf32>)
+    outs(%output: memref<1x1x2xf32>)
+  return
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_max_tensor
 // CHECK:         %{{.+}} = linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x4x1xf32>, tensor<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>
 func.func @pooling_nhwc_max_tensor(%input: tensor<1x4x4x1xf32>) -> tensor<1x2x2x1xf32> {
   %fake = tensor.empty() : tensor<3x3xf32>
   %init = tensor.empty() : tensor<1x2x2x1xf32>
   %cst = arith.constant 0.000000e+00 : f32
   %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>
   %res = linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x4x4x1xf32>, tensor<3x3xf32>)
     outs(%fill: tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>
   return %res : tensor<1x2x2x1xf32>
 }
 
 // -----
+// CHECK-LABEL: func @pooling_nwc_max_tensor
+// CHECK:         %{{.+}} = linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x1xf32>, tensor<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+func.func @pooling_nwc_max_tensor(%input: tensor<1x4x1xf32>) -> tensor<1x2x1xf32> {
+  %fake = tensor.empty() : tensor<3xf32>
+  %init = tensor.empty() : tensor<1x2x1xf32>
+  %cst = arith.constant 0.000000e+00 : f32
+  %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+  %res = linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x4x1xf32>, tensor<3xf32>)
+    outs(%fill: tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+  return %res : tensor<1x2x1xf32>
+}
+
+// -----
 // CHECK-LABEL: func @pooling_nchw_max_tensor
 // CHECK:         %{{.+}} = linalg.pooling_nchw_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x1x4x4xf32>, tensor<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x1x2x2xf32>) -> tensor<1x1x2x2xf32>
 
 func.func @pooling_nchw_max_tensor(%input: tensor<1x1x4x4xf32>) -> tensor<1x1x2x2xf32> {
   %fake = tensor.empty() : tensor<3x3xf32>
   %init = tensor.empty() : tensor<1x1x2x2xf32>
   %cst = arith.constant 0.000000e+00 : f32
   %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x1x2x2xf32>) -> tensor<1x1x2x2xf32>
   %res = linalg.pooling_nchw_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x1x4x4xf32>, tensor<3x3xf32>)
     outs(%fill: tensor<1x1x2x2xf32>) -> tensor<1x1x2x2xf32>
   return %res : tensor<1x1x2x2xf32>
 }
 
 // -----
+// CHECK-LABEL: func @pooling_ncw_max_tensor
+// CHECK:         %{{.+}} = linalg.pooling_ncw_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x1x4xf32>, tensor<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x1x2xf32>) -> tensor<1x1x2xf32>
+
+func.func @pooling_ncw_max_tensor(%input: tensor<1x1x4xf32>) -> tensor<1x1x2xf32> {
+  %fake = tensor.empty() : tensor<3xf32>
+  %init = tensor.empty() : tensor<1x1x2xf32>
+  %cst = arith.constant 0.000000e+00 : f32
+  %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x1x2xf32>) -> tensor<1x1x2xf32>
+  %res = linalg.pooling_ncw_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x1x4xf32>, tensor<3xf32>)
+    outs(%fill: tensor<1x1x2xf32>) -> tensor<1x1x2xf32>
+  return %res : tensor<1x1x2xf32>
+}
+
+// -----
 
 // CHECK-LABEL: func @pooling_nhwc_max
 // CHECK:         linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x4x1xf32>, memref<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : memref<1x2x2x1xf32>)
 func.func @pooling_nhwc_max(%input: memref<1x4x4x1xf32>, %fake: memref<3x3xf32>, %output: memref<1x2x2x1xf32>) {
   linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: memref<1x4x4x1xf32>, memref<3x3xf32>)
     outs(%output: memref<1x2x2x1xf32>)
   return
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_max
+// CHECK:         linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x1xf32>, memref<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : memref<1x2x1xf32>)
+func.func @pooling_nwc_max(%input: memref<1x4x1xf32>, %fake: memref<3xf32>, %output: memref<1x2x1xf32>) {
+  linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: memref<1x4x1xf32>, memref<3xf32>)
+    outs(%output: memref<1x2x1xf32>)
+  return
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_i8_max_tensor
 // CHECK:         %{{.+}} = linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x4x1xi8>, tensor<3x3xi8>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x2x2x1xi8>) -> tensor<1x2x2x1xi8>
 func.func @pooling_nhwc_i8_max_tensor(%input: tensor<1x4x4x1xi8>) -> tensor<1x2x2x1xi8> {
   %fake = tensor.empty() : tensor<3x3xi8>
   %init = tensor.empty() : tensor<1x2x2x1xi8>
   %cst = arith.constant 0 : i8
   %fill = linalg.fill ins(%cst : i8) outs(%init : tensor<1x2x2x1xi8>) -> tensor<1x2x2x1xi8>
   %res = linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x4x4x1xi8>, tensor<3x3xi8>)
     outs(%fill: tensor<1x2x2x1xi8>) -> tensor<1x2x2x1xi8>
   return %res : tensor<1x2x2x1xi8>
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_i8_max_tensor
+// CHECK:         %{{.+}} = linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x1xi8>, tensor<3xi8>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x2x1xi8>) -> tensor<1x2x1xi8>
+func.func @pooling_nwc_i8_max_tensor(%input: tensor<1x4x1xi8>) -> tensor<1x2x1xi8> {
+  %fake = tensor.empty() : tensor<3xi8>
+  %init = tensor.empty() : tensor<1x2x1xi8>
+  %cst = arith.constant 0 : i8
+  %fill = linalg.fill ins(%cst : i8) outs(%init : tensor<1x2x1xi8>) -> tensor<1x2x1xi8>
+  %res = linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x4x1xi8>, tensor<3xi8>)
+    outs(%fill: tensor<1x2x1xi8>) -> tensor<1x2x1xi8>
+  return %res : tensor<1x2x1xi8>
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_i8_max
 // CHECK:         linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x4x1xi8>, memref<3x3xi8>)
 // CHECK-SAME:      outs(%{{.+}} : memref<1x2x2x1xi8>)
 func.func @pooling_nhwc_i8_max(%input: memref<1x4x4x1xi8>, %fake: memref<3x3xi8>, %output: memref<1x2x2x1xi8>) {
   linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: memref<1x4x4x1xi8>, memref<3x3xi8>)
     outs(%output: memref<1x2x2x1xi8>)
   return
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_i8_max
+// CHECK:         linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x1xi8>, memref<3xi8>)
+// CHECK-SAME:      outs(%{{.+}} : memref<1x2x1xi8>)
+func.func @pooling_nwc_i8_max(%input: memref<1x4x1xi8>, %fake: memref<3xi8>, %output: memref<1x2x1xi8>) {
+  linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: memref<1x4x1xi8>, memref<3xi8>)
+    outs(%output: memref<1x2x1xi8>)
+  return
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_i16_max_tensor
 // CHECK:         %{{.+}} = linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x4x1xi16>, tensor<3x3xi16>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x2x2x1xi16>) -> tensor<1x2x2x1xi16>
 func.func @pooling_nhwc_i16_max_tensor(%input: tensor<1x4x4x1xi16>) -> tensor<1x2x2x1xi16> {
   %fake = tensor.empty() : tensor<3x3xi16>
   %init = tensor.empty() : tensor<1x2x2x1xi16>
   %cst = arith.constant 0 : i16
   %fill = linalg.fill ins(%cst : i16) outs(%init : tensor<1x2x2x1xi16>) -> tensor<1x2x2x1xi16>
   %res = linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x4x4x1xi16>, tensor<3x3xi16>)
     outs(%fill: tensor<1x2x2x1xi16>) -> tensor<1x2x2x1xi16>
   return %res : tensor<1x2x2x1xi16>
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_i16_max_tensor
+// CHECK:         %{{.+}} = linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x1xi16>, tensor<3xi16>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x2x1xi16>) -> tensor<1x2x1xi16>
+func.func @pooling_nwc_i16_max_tensor(%input: tensor<1x4x1xi16>) -> tensor<1x2x1xi16> {
+  %fake = tensor.empty() : tensor<3xi16>
+  %init = tensor.empty() : tensor<1x2x1xi16>
+  %cst = arith.constant 0 : i16
+  %fill = linalg.fill ins(%cst : i16) outs(%init : tensor<1x2x1xi16>) -> tensor<1x2x1xi16>
+  %res = linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x4x1xi16>, tensor<3xi16>)
+    outs(%fill: tensor<1x2x1xi16>) -> tensor<1x2x1xi16>
+  return %res : tensor<1x2x1xi16>
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_i16_max
 // CHECK:         linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x4x1xi16>, memref<3x3xi16>)
 // CHECK-SAME:      outs(%{{.+}} : memref<1x2x2x1xi16>)
 func.func @pooling_nhwc_i16_max(%input: memref<1x4x4x1xi16>, %fake: memref<3x3xi16>, %output: memref<1x2x2x1xi16>) {
   linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: memref<1x4x4x1xi16>, memref<3x3xi16>)
     outs(%output: memref<1x2x2x1xi16>)
   return
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_i16_max
+// CHECK:         linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x1xi16>, memref<3xi16>)
+// CHECK-SAME:      outs(%{{.+}} : memref<1x2x1xi16>)
+func.func @pooling_nwc_i16_max(%input: memref<1x4x1xi16>, %fake: memref<3xi16>, %output: memref<1x2x1xi16>) {
+  linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: memref<1x4x1xi16>, memref<3xi16>)
+    outs(%output: memref<1x2x1xi16>)
+  return
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_i32_max_tensor
 // CHECK:         %{{.+}} = linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x4x1xi32>, tensor<3x3xi32>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x2x2x1xi32>) -> tensor<1x2x2x1xi32>
 func.func @pooling_nhwc_i32_max_tensor(%input: tensor<1x4x4x1xi32>) -> tensor<1x2x2x1xi32> {
   %fake = tensor.empty() : tensor<3x3xi32>
   %init = tensor.empty() : tensor<1x2x2x1xi32>
   %cst = arith.constant 0 : i32
   %fill = linalg.fill ins(%cst : i32) outs(%init : tensor<1x2x2x1xi32>) -> tensor<1x2x2x1xi32>
   %res = linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x4x4x1xi32>, tensor<3x3xi32>)
     outs(%fill: tensor<1x2x2x1xi32>) -> tensor<1x2x2x1xi32>
   return %res : tensor<1x2x2x1xi32>
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_i32_max_tensor
+// CHECK:         %{{.+}} = linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x1xi32>, tensor<3xi32>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x2x1xi32>) -> tensor<1x2x1xi32>
+func.func @pooling_nwc_i32_max_tensor(%input: tensor<1x4x1xi32>) -> tensor<1x2x1xi32> {
+  %fake = tensor.empty() : tensor<3xi32>
+  %init = tensor.empty() : tensor<1x2x1xi32>
+  %cst = arith.constant 0 : i32
+  %fill = linalg.fill ins(%cst : i32) outs(%init : tensor<1x2x1xi32>) -> tensor<1x2x1xi32>
+  %res = linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x4x1xi32>, tensor<3xi32>)
+    outs(%fill: tensor<1x2x1xi32>) -> tensor<1x2x1xi32>
+  return %res : tensor<1x2x1xi32>
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_i32_max
 // CHECK:         linalg.pooling_nhwc_max
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x4x1xi32>, memref<3x3xi32>)
 // CHECK-SAME:      outs(%{{.+}} : memref<1x2x2x1xi32>)
 func.func @pooling_nhwc_i32_max(%input: memref<1x4x4x1xi32>, %fake: memref<3x3xi32>, %output: memref<1x2x2x1xi32>) {
   linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: memref<1x4x4x1xi32>, memref<3x3xi32>)
     outs(%output: memref<1x2x2x1xi32>)
   return
 }
 
+// -----
+
+// CHECK-LABEL: func @pooling_nwc_i32_max
+// CHECK:         linalg.pooling_nwc_max
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x1xi32>, memref<3xi32>)
+// CHECK-SAME:      outs(%{{.+}} : memref<1x2x1xi32>)
+func.func @pooling_nwc_i32_max(%input: memref<1x4x1xi32>, %fake: memref<3xi32>, %output: memref<1x2x1xi32>) {
+  linalg.pooling_nwc_max {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: memref<1x4x1xi32>, memref<3xi32>)
+    outs(%output: memref<1x2x1xi32>)
+  return
+}
+
 
 // -----
 
 // CHECK-LABEL: func @pooling_nhwc_min_tensor
 // CHECK:         %{{.+}} = linalg.pooling_nhwc_min
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x4x1xf32>, tensor<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>
 func.func @pooling_nhwc_min_tensor(%input: tensor<1x4x4x1xf32>) -> tensor<1x2x2x1xf32> {
   %fake = tensor.empty() : tensor<3x3xf32>
   %init = tensor.empty() : tensor<1x2x2x1xf32>
   %cst = arith.constant 0.000000e+00 : f32
   %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>
   %res = linalg.pooling_nhwc_min {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: tensor<1x4x4x1xf32>, tensor<3x3xf32>)
     outs(%fill: tensor<1x2x2x1xf32>) -> tensor<1x2x2x1xf32>
   return %res : tensor<1x2x2x1xf32>
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_min_tensor
+// CHECK:         %{{.+}} = linalg.pooling_nwc_min
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x1xf32>, tensor<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+func.func @pooling_nwc_min_tensor(%input: tensor<1x4x1xf32>) -> tensor<1x2x1xf32> {
+  %fake = tensor.empty() : tensor<3xf32>
+  %init = tensor.empty() : tensor<1x2x1xf32>
+  %cst = arith.constant 0.000000e+00 : f32
+  %fill = linalg.fill ins(%cst : f32) outs(%init : tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+  %res = linalg.pooling_nwc_min {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: tensor<1x4x1xf32>, tensor<3xf32>)
+    outs(%fill: tensor<1x2x1xf32>) -> tensor<1x2x1xf32>
+  return %res : tensor<1x2x1xf32>
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_nhwc_min
 // CHECK:         linalg.pooling_nhwc_min
 // CHECK-SAME:      dilations = dense<1> : tensor<2xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<2xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x4x1xf32>, memref<3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : memref<1x2x2x1xf32>)
 func.func @pooling_nhwc_min(%input: memref<1x4x4x1xf32>, %fake: memref<3x3xf32>, %output: memref<1x2x2x1xf32>) {
   linalg.pooling_nhwc_min {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>}
     ins(%input, %fake: memref<1x4x4x1xf32>, memref<3x3xf32>)
     outs(%output: memref<1x2x2x1xf32>)
   return
 }
 
 // -----
 
+// CHECK-LABEL: func @pooling_nwc_min
+// CHECK:         linalg.pooling_nwc_min
+// CHECK-SAME:      dilations = dense<1> : tensor<1xi64>
+// CHECK-SAME:      strides = dense<1> : tensor<1xi64>
+// CHECK-SAME:      ins(%{{.+}}, %{{.+}} : memref<1x4x1xf32>, memref<3xf32>)
+// CHECK-SAME:      outs(%{{.+}} : memref<1x2x1xf32>)
+func.func @pooling_nwc_min(%input: memref<1x4x1xf32>, %fake: memref<3xf32>, %output: memref<1x2x1xf32>) {
+  linalg.pooling_nwc_min {dilations = dense<1> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %fake: memref<1x4x1xf32>, memref<3xf32>)
+    outs(%output: memref<1x2x1xf32>)
+  return
+}
+
+// -----
+
 // CHECK-LABEL: func @pooling_ndhwc_sum_tensor
 // CHECK:         %{{.+}} = linalg.pooling_ndhwc_sum
 // CHECK-SAME:      dilations = dense<1> : tensor<3xi64>
 // CHECK-SAME:      strides = dense<1> : tensor<3xi64>
 // CHECK-SAME:      ins(%{{.+}}, %{{.+}} : tensor<1x4x4x4x1xf32>, tensor<3x3x3xf32>)
 // CHECK-SAME:      outs(%{{.+}} : tensor<1x2x2x2x1xf32>) -> tensor<1x2x2x2x1xf32>
 func.func @pooling_ndhwc_sum_tensor(%input: tensor<1x4x4x4x1xf32>) -> tensor<1x2x2x2x1xf32> {
   %fake = tensor.empty() : tensor<3x3x3xf32>

mlir/test/Dialect/Linalg/transform-op-decompose.mlir

func.func @depthwise_conv_2d_nhwc_hwc(%input: tensor<1x1x113x96xf32>, %filter: tensor<1x3x96xf32>) -> tensor<1x1x56x96xf32> {
   // CHECK: %[[INSERTED:.+]] = tensor.insert_slice %[[OPRES]] into %[[RES]]
   %0 = linalg.depthwise_conv_2d_nhwc_hwc {dilations = dense<1> : vector<2xi64>, strides = dense<2> : vector<2xi64>}
          ins(%input, %filter: tensor<1x1x113x96xf32>, tensor<1x3x96xf32>)
          outs(%init: tensor<1x1x56x96xf32>) -> tensor<1x1x56x96xf32>
   // CHECK: %[[INSERTED]]
   return %0: tensor<1x1x56x96xf32>
 }
 
+// CHECK-LABEL: @pooling_nhwc_sum
+// CHECK-SAME: %[[ARG0:.+]]: tensor<?x1x?x?xf32>,
+// CHECK-SAME: %[[ARG1:.+]]: tensor<1x?xf32>
+// CHECK-SAME: %[[ARG2:.+]]: tensor<?x1x?x?xf32>
+func.func @pooling_nhwc_sum(%input: tensor<?x1x?x?xf32>, %filter: tensor<1x?xf32>, %init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32> {
+  // CHECK: %[[SLICE0:.+]] = tensor.extract_slice %[[ARG0]]
+  // CHECK: %[[SLICE1:.+]] = tensor.extract_slice %[[ARG1]]
+  // CHECK: %[[SLICE2:.+]] = tensor.extract_slice %[[ARG2]]
+  // CHECK: %[[SLICERES:.+]] = linalg.pooling_nwc_sum
+  // CHECK: %[[RES:.+]] = tensor.insert_slice %[[SLICERES]] into %[[ARG2]]
+  %0 = linalg.pooling_nhwc_sum {dilations = dense<1> : tensor<2xi64>,
+                                strides = dense<1> : tensor<2xi64>}
+     ins (%input, %filter: tensor<?x1x?x?xf32>, tensor<1x?xf32>)
+    outs (%init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32>
+  // CHECK: return %[[RES]]
+  return %0 : tensor<?x1x?x?xf32>
+}
+
+// CHECK-LABEL: @pooling_nchw_sum
+// CHECK-SAME: (%[[ARG0:[0-9a-z]+]]: tensor<?x?x1x?xf32>,
+// CHECK-SAME: %[[ARG1:[0-9a-z]+]]: tensor<1x?xf32>,
+// CHECK-SAME: %[[ARG2:[0-9a-z]+]]: tensor<?x?x1x?xf32>)
+func.func @pooling_nchw_sum(%input: tensor<?x?x1x?xf32>, %filter: tensor<1x?xf32>, %init: tensor<?x?x1x?xf32>) -> tensor<?x?x1x?xf32> {
+  // CHECK: %[[SLICE0:.+]] = tensor.extract_slice %[[ARG0]]
+  // CHECK: %[[SLICE1:.+]] = tensor.extract_slice %[[ARG1]]
+  // CHECK: %[[SLICE2:.+]] = tensor.extract_slice %[[ARG2]]
+  // CHECK: %[[SLICERES:.+]] = linalg.pooling_ncw_sum
+  // CHECK: %[[RES:.+]] = tensor.insert_slice %[[SLICERES]] into %[[ARG2]]
+  %0 = linalg.pooling_nchw_sum {dilations = dense<1> : tensor<2xi64>,
+                                strides = dense<1> : tensor<2xi64>}
+     ins (%input, %filter: tensor<?x?x1x?xf32>, tensor<1x?xf32>)
+    outs (%init: tensor<?x?x1x?xf32>) -> tensor<?x?x1x?xf32>
+  // CHECK: return %[[RES]]
+  return %0 : tensor<?x?x1x?xf32>
+}
+
+// CHECK-LABEL: @pooling_nhwc_max
+// CHECK-SAME: %[[ARG0:.+]]: tensor<?x1x?x?xf32>,
+// CHECK-SAME: %[[ARG1:.+]]: tensor<1x?xf32>
+// CHECK-SAME: %[[ARG2:.+]]: tensor<?x1x?x?xf32>
+func.func @pooling_nhwc_max(%input: tensor<?x1x?x?xf32>, %filter: tensor<1x?xf32>, %init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32> {
+  // CHECK: %[[SLICE0:.+]] = tensor.extract_slice %[[ARG0]]
+  // CHECK: %[[SLICE1:.+]] = tensor.extract_slice %[[ARG1]]
+  // CHECK: %[[SLICE2:.+]] = tensor.extract_slice %[[ARG2]]
+  // CHECK: %[[SLICERES:.+]] = linalg.pooling_nwc_max
+  // CHECK: %[[RES:.+]] = tensor.insert_slice %[[SLICERES]] into %[[ARG2]]
+  %0 = linalg.pooling_nhwc_max {dilations = dense<1> : tensor<2xi64>,
+                                strides = dense<1> : tensor<2xi64>}
+     ins (%input, %filter: tensor<?x1x?x?xf32>, tensor<1x?xf32>)
+    outs (%init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32>
+  // CHECK: return %[[RES]]
+  return %0 : tensor<?x1x?x?xf32>
+}
+
+// CHECK-LABEL: @pooling_nhwc_max_unsigned
+// CHECK-SAME: %[[ARG0:.+]]: tensor<?x1x?x?xf32>,
+// CHECK-SAME: %[[ARG1:.+]]: tensor<1x?xf32>
+// CHECK-SAME: %[[ARG2:.+]]: tensor<?x1x?x?xf32>
+func.func @pooling_nhwc_max_unsigned(%input: tensor<?x1x?x?xf32>, %filter: tensor<1x?xf32>, %init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32> {
+  // CHECK: %[[SLICE0:.+]] = tensor.extract_slice %[[ARG0]]
+  // CHECK: %[[SLICE1:.+]] = tensor.extract_slice %[[ARG1]]
+  // CHECK: %[[SLICE2:.+]] = tensor.extract_slice %[[ARG2]]
+  // CHECK: %[[SLICERES:.+]] = linalg.pooling_nwc_max_unsigned
+  // CHECK: %[[RES:.+]] = tensor.insert_slice %[[SLICERES]] into %[[ARG2]]
+  %0 = linalg.pooling_nhwc_max_unsigned {dilations = dense<1> : tensor<2xi64>,
+                                strides = dense<1> : tensor<2xi64>}
+     ins (%input, %filter: tensor<?x1x?x?xf32>, tensor<1x?xf32>)
+    outs (%init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32>
+  // CHECK: return %[[RES]]
+  return %0 : tensor<?x1x?x?xf32>
+}
+
+// CHECK-LABEL: @pooling_nhwc_min
+// CHECK-SAME: %[[ARG0:.+]]: tensor<?x1x?x?xf32>,
+// CHECK-SAME: %[[ARG1:.+]]: tensor<1x?xf32>
+// CHECK-SAME: %[[ARG2:.+]]: tensor<?x1x?x?xf32>
+func.func @pooling_nhwc_min(%input: tensor<?x1x?x?xf32>, %filter: tensor<1x?xf32>, %init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32> {
+  // CHECK: %[[SLICE0:.+]] = tensor.extract_slice %[[ARG0]]
+  // CHECK: %[[SLICE1:.+]] = tensor.extract_slice %[[ARG1]]
+  // CHECK: %[[SLICE2:.+]] = tensor.extract_slice %[[ARG2]]
+  // CHECK: %[[SLICERES:.+]] = linalg.pooling_nwc_min
+  // CHECK: %[[RES:.+]] = tensor.insert_slice %[[SLICERES]] into %[[ARG2]]
+  %0 = linalg.pooling_nhwc_min {dilations = dense<1> : tensor<2xi64>,
+                                strides = dense<1> : tensor<2xi64>}
+     ins (%input, %filter: tensor<?x1x?x?xf32>, tensor<1x?xf32>)
+    outs (%init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32>
+  // CHECK: return %[[RES]]
+  return %0 : tensor<?x1x?x?xf32>
+}
+
+// CHECK-LABEL: @pooling_nhwc_min_unsigned
+// CHECK-SAME: %[[ARG0:.+]]: tensor<?x1x?x?xf32>,
+// CHECK-SAME: %[[ARG1:.+]]: tensor<1x?xf32>
+// CHECK-SAME: %[[ARG2:.+]]: tensor<?x1x?x?xf32>
+func.func @pooling_nhwc_min_unsigned(%input: tensor<?x1x?x?xf32>, %filter: tensor<1x?xf32>, %init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32> {
+  // CHECK: %[[SLICE0:.+]] = tensor.extract_slice %[[ARG0]]
+  // CHECK: %[[SLICE1:.+]] = tensor.extract_slice %[[ARG1]]
+  // CHECK: %[[SLICE2:.+]] = tensor.extract_slice %[[ARG2]]
+  // CHECK: %[[SLICERES:.+]] = linalg.pooling_nwc_min_unsigned
+  // CHECK: %[[RES:.+]] = tensor.insert_slice %[[SLICERES]] into %[[ARG2]]
+  %0 = linalg.pooling_nhwc_min_unsigned {dilations = dense<1> : tensor<2xi64>,
+                                strides = dense<1> : tensor<2xi64>}
+     ins (%input, %filter: tensor<?x1x?x?xf32>, tensor<1x?xf32>)
+    outs (%init: tensor<?x1x?x?xf32>) -> tensor<?x1x?x?xf32>
+  // CHECK: return %[[RES]]
+  return %0 : tensor<?x1x?x?xf32>
+}
+
+// CHECK-LABEL: @pooling_nchw_max
+// CHECK-SAME: (%[[ARG0:[0-9a-z]+]]: tensor<?x?x1x?xf32>,
+// CHECK-SAME: %[[ARG1:[0-9a-z]+]]: tensor<1x?xf32>,
+// CHECK-SAME: %[[ARG2:[0-9a-z]+]]: tensor<?x?x1x?xf32>)
+func.func @pooling_nchw_max(%input: tensor<?x?x1x?xf32>, %filter: tensor<1x?xf32>, %init: tensor<?x?x1x?xf32>) -> tensor<?x?x1x?xf32> {
+  // CHECK: %[[SLICE0:.+]] = tensor.extract_slice %[[ARG0]]
+  // CHECK: %[[SLICE1:.+]] = tensor.extract_slice %[[ARG1]]
+  // CHECK: %[[SLICE2:.+]] = tensor.extract_slice %[[ARG2]]
+  // CHECK: %[[SLICERES:.+]] = linalg.pooling_ncw_max
+  // CHECK: %[[RES:.+]] = tensor.insert_slice %[[SLICERES]] into %[[ARG2]]
+  %0 = linalg.pooling_nchw_max {dilations = dense<1> : tensor<2xi64>,
+                                strides = dense<1> : tensor<2xi64>}
+     ins (%input, %filter: tensor<?x?x1x?xf32>, tensor<1x?xf32>)
+    outs (%init: tensor<?x?x1x?xf32>) -> tensor<?x?x1x?xf32>
+  // CHECK: return %[[RES]]
+  return %0 : tensor<?x?x1x?xf32>
+}
+
 transform.sequence failures(propagate) {
 ^bb1(%arg1: !pdl.operation):
   %0 = transform.structured.match interface{LinalgOp} in %arg1
   %1 = transform.structured.decompose %0
 }

mlir/test/Dialect/Linalg/vectorize-convolution.mlir

 /// Read the whole data in one shot.
 //      CHECK:   %[[V_INPUT_R:.+]] = vector.transfer_read %[[INPUT]][%[[C0]], %[[C0]], %[[C0]]]
 //      CHECK:   %[[V_FILTER_R:.+]] = vector.transfer_read %[[FILTER]][%[[C0]], %[[C0]], %[[C0]]]
 //      CHECK:   %[[V_OUTPUT_R:.+]] = vector.transfer_read %[[OUTPUT]][%[[C0]], %[[C0]], %[[C0]]]
 //      CHECK:   %[[V_FILTER_1:.+]] = vector.extract %[[V_FILTER_R]][0] : vector<1x3x2xf16>
 //      CHECK:   %[[CONT:.*]] = vector.contract
 //        {{.*}} %[[V_INPUT_R]], %[[V_FILTER_1]], %[[V_OUTPUT_R]] : vector<1x2x3xf16>, vector<3x2xf16> into vector<1x2x2xf32>
 //      CHECK:   vector.transfer_write %[[CONT]], %[[OUTPUT]][%[[C0]], %[[C0]], %[[C0]]]
+
+
+
+func.func @pooling_nwc_sum_memref_1_2_1_3(%input: memref<4x4x3xf32>, %filter: memref<1xf32>, %output: memref<4x2x3xf32>) {
+  linalg.pooling_nwc_sum
+    {dilations = dense<1> : tensor<1xi64>, strides = dense<3> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x4x3xf32>, memref<1xf32>)
+    outs(%output : memref<4x2x3xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_nwc_sum_memref_1_2_1_3
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x4x3xf32>, %[[Varg1:.+]]: memref<1xf32>, %[[Varg2:.+]]: memref<4x2x3xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x4x3xf32>, vector<4x4x3xf32>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x2x3xf32>, vector<4x2x3xf32>
+// CHECK-DAG: %[[V2:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V3:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 3, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 1, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V6:.+]] = arith.addf %[[V2]], %[[V4]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V7:.+]] = arith.addf %[[V3]], %[[V5]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V8:.+]] = vector.insert_strided_slice %[[V6]], %[[V1]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: %[[V9:.+]] = vector.insert_strided_slice %[[V7]], %[[V8]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: vector.transfer_write %[[V9]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x2x3xf32>, memref<4x2x3xf32>
+
+// -----
+
+func.func @pooling_nwc_max_memref_1_2_1_3(%input: memref<4x4x3xf32>, %filter: memref<1xf32>, %output: memref<4x2x3xf32>) {
+  linalg.pooling_nwc_max
+    {dilations = dense<1> : tensor<1xi64>, strides = dense<3> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x4x3xf32>, memref<1xf32>)
+    outs(%output : memref<4x2x3xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_nwc_max_memref_1_2_1_3
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x4x3xf32>, %[[Varg1:.+]]: memref<1xf32>, %[[Varg2:.+]]: memref<4x2x3xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x4x3xf32>, vector<4x4x3xf32>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x2x3xf32>, vector<4x2x3xf32>
+// CHECK-DAG: %[[V2:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V3:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 3, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 1, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V6:.+]] = arith.maxf %[[V2]], %[[V4]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V7:.+]] = arith.maxf %[[V3]], %[[V5]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V8:.+]] = vector.insert_strided_slice %[[V6]], %[[V1]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: %[[V9:.+]] = vector.insert_strided_slice %[[V7]], %[[V8]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: vector.transfer_write %[[V9]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x2x3xf32>, memref<4x2x3xf32>
+
+// -----
+
+// The i8i8i32 case is similar to f32 case, so checking one case is enough for
+// test coverage.
+func.func @pooling_nwc_sum_i8i8i32_memref_1_2_1_3(%input: memref<4x4x3xi8>, %filter: memref<1xi8>, %output: memref<4x2x3xi32>) {
+  linalg.pooling_nwc_sum
+    {dilations = dense<1> : tensor<1xi64>, strides = dense<3> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x4x3xi8>, memref<1xi8>)
+    outs(%output : memref<4x2x3xi32>)
+  return
+}
+
+// CHECK: func.func @pooling_nwc_sum_i8i8i32_memref_1_2_1_3
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x4x3xi8>, %[[Varg1:.+]]: memref<1xi8>, %[[Varg2:.+]]: memref<4x2x3xi32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vc0_i8:.+]] = arith.constant 0 : i8
+// CHECK-DAG: %[[Vc0_i32:.+]] = arith.constant 0 : i32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vc0_i8]] {in_bounds = [true, true, true]} : memref<4x4x3xi8>, vector<4x4x3xi8>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vc0_i32]] {in_bounds = [true, true, true]} : memref<4x2x3xi32>, vector<4x2x3xi32>
+// CHECK-DAG: %[[V2:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xi8> to vector<4x1x3xi8>
+// CHECK-DAG: %[[V3:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 3, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xi8> to vector<4x1x3xi8>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xi32> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 1, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xi32> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V6:.+]] = arith.extsi %[[V2]] : vector<4x1x3xi8> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V7:.+]] = arith.addi %[[V6]], %[[V4]] : vector<4x1x3xi32>
+// CHECK-DAG: %[[V8:.+]] = arith.extsi %[[V3]] : vector<4x1x3xi8> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V9:.+]] = arith.addi %[[V8]], %[[V5]] : vector<4x1x3xi32>
+// CHECK-DAG: %[[V10:.+]] = vector.insert_strided_slice %[[V7]], %[[V1]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xi32> into vector<4x2x3xi32>
+// CHECK-DAG: %[[V11:.+]] = vector.insert_strided_slice %[[V9]], %[[V10]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xi32> into vector<4x2x3xi32>
+// CHECK-DAG: vector.transfer_write %[[V11]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x2x3xi32>, memref<4x2x3xi32>
+// CHECK-DAG: return
+
+// -----
+
+// The i8i8i32 case is similar to f32 case, so checking one case is enough for
+// test coverage.
+func.func @pooling_nwc_max_i8i8i32_memref_1_2_1_3(%input: memref<4x4x3xi8>, %filter: memref<1xi8>, %output: memref<4x2x3xi32>) {
+  linalg.pooling_nwc_max
+    {dilations = dense<1> : tensor<1xi64>, strides = dense<3> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x4x3xi8>, memref<1xi8>)
+    outs(%output : memref<4x2x3xi32>)
+  return
+}
+
+// CHECK: func.func @pooling_nwc_max_i8i8i32_memref_1_2_1_3
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x4x3xi8>, %[[Varg1:.+]]: memref<1xi8>, %[[Varg2:.+]]: memref<4x2x3xi32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vc0_i8:.+]] = arith.constant 0 : i8
+// CHECK-DAG: %[[Vc0_i32:.+]] = arith.constant 0 : i32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vc0_i8]] {in_bounds = [true, true, true]} : memref<4x4x3xi8>, vector<4x4x3xi8>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vc0_i32]] {in_bounds = [true, true, true]} : memref<4x2x3xi32>, vector<4x2x3xi32>
+// CHECK-DAG: %[[V2:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xi8> to vector<4x1x3xi8>
+// CHECK-DAG: %[[V3:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 3, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xi8> to vector<4x1x3xi8>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xi32> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 1, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xi32> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V6:.+]] = arith.extsi %[[V2]] : vector<4x1x3xi8> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V7:.+]] = arith.maxsi %[[V6]], %[[V4]] : vector<4x1x3xi32>
+// CHECK-DAG: %[[V8:.+]] = arith.extsi %[[V3]] : vector<4x1x3xi8> to vector<4x1x3xi32>
+// CHECK-DAG: %[[V9:.+]] = arith.maxsi %[[V8]], %[[V5]] : vector<4x1x3xi32>
+// CHECK-DAG: %[[V10:.+]] = vector.insert_strided_slice %[[V7]], %[[V1]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xi32> into vector<4x2x3xi32>
+// CHECK-DAG: %[[V11:.+]] = vector.insert_strided_slice %[[V9]], %[[V10]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xi32> into vector<4x2x3xi32>
+// CHECK-DAG: vector.transfer_write %[[V11]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x2x3xi32>, memref<4x2x3xi32>
+// CHECK-DAG: return
+
+// -----
+
+func.func @pooling_nwc_sum_memref_2_2_2_3(%input: memref<4x6x3xf32>, %filter: memref<2xf32>, %output: memref<4x2x3xf32>) {
+  linalg.pooling_nwc_sum
+    {dilations = dense<2> : tensor<1xi64>, strides = dense<3> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x6x3xf32>, memref<2xf32>)
+    outs(%output : memref<4x2x3xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_nwc_sum_memref_2_2_2_3
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x6x3xf32>, %[[Varg1:.+]]: memref<2xf32>, %[[Varg2:.+]]: memref<4x2x3xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x6x3xf32>, vector<4x6x3xf32>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x2x3xf32>, vector<4x2x3xf32>
+// CHECK-DAG: %[[V2:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V3:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 3, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 2, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 5, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V6:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V7:.+]] = vector.extract_strided_slice %[[V1]] {offsets = [0, 1, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V8:.+]] = arith.addf %[[V2]], %[[V6]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V9:.+]] = arith.addf %[[V3]], %[[V7]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V10:.+]] = arith.addf %[[V4]], %[[V8]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V11:.+]] = arith.addf %[[V5]], %[[V9]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V12:.+]] = vector.insert_strided_slice %[[V10]], %[[V1]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: %[[V13:.+]] = vector.insert_strided_slice %[[V11]], %[[V12]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: vector.transfer_write %[[V13:.+]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x2x3xf32>, memref<4x2x3xf32>
+
+
+// -----
+
+func.func @pooling_ncw_sum_memref_1_2_1_3(%input: memref<4x3x4xf32>, %filter: memref<1xf32>, %output: memref<4x3x2xf32>) {
+  linalg.pooling_ncw_sum
+    {dilations = dense<1> : tensor<1xi64>, strides = dense<3> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x3x4xf32>, memref<1xf32>)
+    outs(%output : memref<4x3x2xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_ncw_sum_memref_1_2_1_3
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x3x4xf32>, %[[Varg1:.+]]: memref<1xf32>, %[[Varg2:.+]]: memref<4x3x2xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x3x4xf32>, vector<4x3x4xf32>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x3x2xf32>, vector<4x3x2xf32>
+// CHECK-DAG: %[[V2:.+]] = vector.transpose %[[V0]], [0, 2, 1] : vector<4x3x4xf32> to vector<4x4x3xf32>
+// CHECK-DAG: %[[V3:.+]] = vector.transpose %[[V1]], [0, 2, 1] : vector<4x3x2xf32> to vector<4x2x3xf32>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 3, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V6:.+]] = vector.extract_strided_slice %[[V3]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V7:.+]] = vector.extract_strided_slice %[[V3]] {offsets = [0, 1, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V8:.+]] = arith.addf %[[V4]], %[[V6]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V9:.+]] = arith.addf %[[V5]], %[[V7]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V10:.+]] = vector.insert_strided_slice %[[V8]], %[[V3]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: %[[V11:.+]] = vector.insert_strided_slice %[[V9]], %[[V10]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: %[[V12:.+]] = vector.transpose %[[V11]], [0, 2, 1] : vector<4x2x3xf32> to vector<4x3x2xf32>
+// CHECK-DAG: vector.transfer_write %[[V12:.+]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x3x2xf32>, memref<4x3x2xf32>
+
+
+// -----
+
+func.func @pooling_nwc_sum_mixed_type_memref_1_2_1_1(%input: memref<1x2x3xf16>, %filter: memref<1xf16>, %output: memref<1x2x3xf32>) {
+  linalg.pooling_nwc_sum
+  {dilations = dense<1> : vector<1xi64>, strides = dense<1> : vector<1xi64>}
+   ins(%input, %filter : memref<1x2x3xf16>, memref<1xf16>)
+   outs(%output : memref<1x2x3xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_nwc_sum_mixed_type_memref_1_2_1_1
+// CHECK-SAME: (%[[Varg0:.+]]: memref<1x2x3xf16>, %[[Varg1:.+]]: memref<1xf16>, %[[Varg2:.+]]: memref<1x2x3xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f16
+// CHECK-DAG: %[[Vcst_0:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<1x2x3xf16>, vector<1x2x3xf16>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst_0]] {in_bounds = [true, true, true]} : memref<1x2x3xf32>, vector<1x2x3xf32>
+// CHECK-DAG: %[[V2:.+]] = arith.extf %[[V0]] : vector<1x2x3xf16> to vector<1x2x3xf32>
+// CHECK-DAG: %[[V3:.+]] = arith.addf %[[V2]], %[[V1]] : vector<1x2x3xf32>
+// CHECK-DAG: vector.transfer_write %[[V3:.+]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<1x2x3xf32>, memref<1x2x3xf32>
+
+// -----
+
+func.func @pooling_nwc_sum_memref_2_2_2_1(%input: memref<4x4x3xf32>, %filter: memref<2xf32>, %output: memref<4x2x3xf32>) {
+  linalg.pooling_nwc_sum
+    {dilations = dense<2> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x4x3xf32>, memref<2xf32>)
+    outs(%output : memref<4x2x3xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_nwc_sum_memref_2_2_2_1
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x4x3xf32>, %[[Varg1:.+]]: memref<2xf32>, %[[Varg2:.+]]: memref<4x2x3xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x4x3xf32>, vector<4x4x3xf32>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x2x3xf32>, vector<4x2x3xf32>
+// CHECK-DAG: %[[V2:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 0, 0], sizes = [4, 2, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x2x3xf32>
+// CHECK-DAG: %[[V3:.+]] = vector.extract_strided_slice %[[V0]] {offsets = [0, 2, 0], sizes = [4, 2, 3], strides = [1, 1, 1]} : vector<4x4x3xf32> to vector<4x2x3xf32>
+// CHECK-DAG: %[[V4:.+]] = arith.addf %[[V2]], %[[V1]] : vector<4x2x3xf32>
+// CHECK-DAG: %[[V5:.+]] = arith.addf %[[V3]], %[[V4]] : vector<4x2x3xf32>
+// CHECK-DAG: vector.transfer_write %[[V5:.+]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x2x3xf32>, memref<4x2x3xf32>
+
+
+// -----
+
+func.func @pooling_ncw_sum_memref_2_2_2_3(%input: memref<4x3x6xf32>, %filter: memref<2xf32>, %output: memref<4x3x2xf32>) {
+  linalg.pooling_ncw_sum
+    {dilations = dense<2> : tensor<1xi64>, strides = dense<3> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x3x6xf32>, memref<2xf32>)
+    outs(%output : memref<4x3x2xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_ncw_sum_memref_2_2_2_3
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x3x6xf32>, %[[Varg1:.+]]: memref<2xf32>, %[[Varg2:.+]]: memref<4x3x2xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x3x6xf32>, vector<4x3x6xf32>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x3x2xf32>, vector<4x3x2xf32>
+// CHECK-DAG: %[[V2:.+]] = vector.transpose %[[V0]], [0, 2, 1] : vector<4x3x6xf32> to vector<4x6x3xf32>
+// CHECK-DAG: %[[V3:.+]] = vector.transpose %[[V1]], [0, 2, 1] : vector<4x3x2xf32> to vector<4x2x3xf32>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 3, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V6:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 2, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V7:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 5, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x6x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V8:.+]] = vector.extract_strided_slice %[[V3]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V9:.+]] = vector.extract_strided_slice %[[V3]] {offsets = [0, 1, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
+// CHECK-DAG: %[[V10:.+]] = arith.addf %[[V4]], %[[V8]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V11:.+]] = arith.addf %[[V5]], %[[V9]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V12:.+]] = arith.addf %[[V6]], %[[V10]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V13:.+]] = arith.addf %[[V7]], %[[V11]] : vector<4x1x3xf32>
+// CHECK-DAG: %[[V14:.+]] = vector.insert_strided_slice %[[V12]], %[[V3]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: %[[V15:.+]] = vector.insert_strided_slice %[[V13]], %[[V14]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
+// CHECK-DAG: %[[V16:.+]] = vector.transpose %[[V15]], [0, 2, 1] : vector<4x2x3xf32> to vector<4x3x2xf32>
+// CHECK-DAG: vector.transfer_write %[[V16:.+]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x3x2xf32>, memref<4x3x2xf32>
+
+// -----
+
+func.func @pooling_ncw_sum_memref_2_3_2_1(%input: memref<4x2x5xf32>, %filter: memref<2xf32>, %output: memref<4x2x3xf32>) {
+  linalg.pooling_ncw_sum
+    {dilations = dense<2> : tensor<1xi64>, strides = dense<1> : tensor<1xi64>}
+    ins(%input, %filter : memref<4x2x5xf32>, memref<2xf32>)
+    outs(%output : memref<4x2x3xf32>)
+  return
+}
+
+// CHECK: func.func @pooling_ncw_sum_memref_2_3_2_1
+// CHECK-SAME: (%[[Varg0:.+]]: memref<4x2x5xf32>, %[[Varg1:.+]]: memref<2xf32>, %[[Varg2:.+]]: memref<4x2x3xf32>)
+// CHECK-DAG: %[[Vc0:.+]] = arith.constant 0 : index
+// CHECK-DAG: %[[Vcst:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[V0:.+]] = vector.transfer_read %[[Varg0]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x2x5xf32>, vector<4x2x5xf32>
+// CHECK-DAG: %[[V1:.+]] = vector.transfer_read %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]], %[[Vcst]] {in_bounds = [true, true, true]} : memref<4x2x3xf32>, vector<4x2x3xf32>
+// CHECK-DAG: %[[V2:.+]] = vector.transpose %[[V0]], [0, 2, 1] : vector<4x2x5xf32> to vector<4x5x2xf32>
+// CHECK-DAG: %[[V3:.+]] = vector.transpose %[[V1]], [0, 2, 1] : vector<4x2x3xf32> to vector<4x3x2xf32>
+// CHECK-DAG: %[[V4:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 0, 0], sizes = [4, 3, 2], strides = [1, 1, 1]} : vector<4x5x2xf32> to vector<4x3x2xf32>
+// CHECK-DAG: %[[V5:.+]] = vector.extract_strided_slice %[[V2]] {offsets = [0, 2, 0], sizes = [4, 3, 2], strides = [1, 1, 1]} : vector<4x5x2xf32> to vector<4x3x2xf32>
+// CHECK-DAG: %[[V6:.+]] = arith.addf %[[V4]], %[[V3]] : vector<4x3x2xf32>
+// CHECK-DAG: %[[V7:.+]] = arith.addf %[[V5]], %[[V6]] : vector<4x3x2xf32>
+// CHECK-DAG: %[[V8:.+]] = vector.transpose %[[V7]], [0, 2, 1] : vector<4x3x2xf32> to vector<4x2x3xf32>
+// CHECK-DAG: vector.transfer_write %[[V8:.+]], %[[Varg2]][%[[Vc0]], %[[Vc0]], %[[Vc0]]] {in_bounds = [true, true, true]} : vector<4x2x3xf32>, memref<4x2x3xf32>