%0 = "tf.Conv2D"(%arg0, %arg1) {data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true} : (tensor<?x?x?x96xf32>, tensor<3x3x96x?xf32>) -> tensor<?x?x?x?xf32>
  func.return %0 : tensor<?x?x?x?xf32>
// CHECK: tf.Conv2D
}

// CHECK-LABEL: conv2d_f16

  // CHECK: %[[CONV:.*]] = "tf.Conv2D"(%arg0, %[[CST]]) <{data_format = "NHWC", dilations = [1, 2, 3, 1], explicit_paddings = [], padding = "SAME", strides = [1, 4, 5, 1], use_cudnn_on_gpu = true}>
  // CHECK: return %[[CONV]] : tensor<1x28x23x2xf32>
}

// CHECK-LABEL: @notfuseMulIntoConv2d
// filter and multiply are not broadcastable

// CHECK:           %[[VAL_10:.*]] = "tf.Conv2D"(%[[VAL_5]], %[[VAL_9]]) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true}> : (tensor<16x32x1x256xbf16>, tensor<1x1x256x256xbf16>) -> tensor<16x32x1x256xbf16>
// CHECK:           %[[VAL_11:.*]] = "tf.Const"() <{value = dense<[0, 1, 3, 2]> : tensor<4xi64>}> : () -> tensor<4xi64>

    %0 = "tf.Conv2D"(%arg0, %arg1) {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    %1 = "tf.Relu6"(%0) : (tensor<*xf32>) -> tensor<*xf32>

    }
  }
  attr {
    key: "strides"
    value {
      list {
        i: 1
        i: 2
        i: 2
        i: 1
      }
    }
  }
  attr {
    key: "use_cudnn_on_gpu"
    value {
      b: true
    }
  }
}
node {
  name: "MobilenetV1/Conv2d_0/BatchNorm/gamma"
  op: "Const"
  attr {
    key: "dtype"
    value {
      type: DT_FLOAT
    }

    }
  }
  attr {
    key: "strides"
    value {
      list {
        i: 1
        i: 2
        i: 2
        i: 1
      }
    }
  }
  attr {
    key: "use_cudnn_on_gpu"
    value {
      b: true
    }
  }
}
node {
  name: "MobilenetV1/Conv2d_0/BatchNorm/gamma"
  op: "Const"
  attr {
    key: "dtype"
    value {
      type: DT_FLOAT
    }

    } : (tensor<1x300x300x40xi8>, tensor<3x3x40x40xi8>, tensor<40xf32>, tensor<0xi8>, tensor<f32>, tensor<f32>) -> tensor<1x300x300x40xi8>
  func.return %conv2d : tensor<1x300x300x40xi8>
}

    %3 = "tf.Sub"(%2, %arg6) : (tensor<2x3x3x2xi32>, tensor<i32>) -> tensor<2x3x3x2xi32>
    %4 = "tf.Conv2D"(%1, %3) {dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xi32>, tensor<2x3x3x2xi32>) -> tensor<1x3x2x2xi32>
    %5 = "tf.AddV2"(%4, %arg2) : (tensor<1x3x2x2xi32>, tensor<2xi32>) -> tensor<1x3x2x2xi32>

  %7 = "tf.Conv2D"(%6, %4) {data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true} : (tensor<4x960x1x64xf32>, tensor<3x1x64x128xf32>) -> tensor<4x958x1x128xf32>
  %8 = "tf.Squeeze"(%7) {device = "", squeeze_dims = [2]} : (tensor<4x958x1x128xf32>) -> tensor<4x958x128xf32>

                             (TFL_MatrixDiagOp $diagonal)>;

def LegalizeConv2DBackpropInput : Pat<
  (TF_Conv2DBackpropInputOp $input_sizes, $filter, $out_backprop,
     IsIntList1XY1:$strides,
     BoolAttr:$use_cudnn_on_gpu,
     IsSameOrValid:$padding,
     I64ArrayAttr:$explicit_paddings,
     IsDataFormatNHWC:$data_format,
     IsAllOnes:$dilations),
  (TFL_TransposeConvOp $input_sizes,
     (TFL_TransposeOp $filter,