// CHECK-DAG: %[[cst:.*]] = arith.constant dense<[0, 2, 1, 3]> : tensor<4xi32>
  // CHECK-DAG: %[[cst_1:.*]] = arith.constant dense<[0, 2, 3, 1]> : tensor<4xi32>
  // CHECK: %[[v0:.*]] = "tf.Transpose"(%arg0, %[[cst]]) : (tensor<2x5x7x3xf32>, tensor<4xi32>) -> tensor<2x7x5x3xf32>
  // CHECK: %[[v1:.*]] = "tf.Transpose"(%arg1, %[[cst_1]]) : (tensor<2x4x7x3xf32>, tensor<4xi32>) -> tensor<2x7x3x4xf32>

  func.return %0: tensor<2xi32>
}

// CHECK-LABEL: func @shift_left
func.func @shift_left(%arg0: tensor<4xi32>, %arg1: tensor<4xi32>) -> tensor<4xi32> {
  // CHECK:  mhlo.shift_left %arg0, %arg1 : tensor<4xi32>
  %0 = "tf.LeftShift"(%arg0, %arg1) : (tensor<4xi32>, tensor<4xi32>) -> tensor<4xi32>
  func.return %0 : tensor<4xi32>
}

// CHECK-LABEL: func @div_unranked

        %cst_0 = "tf.Const"() {device = "", value = dense<[3, 4, 5, 6]> : tensor<4xi32>} : () -> tensor<4xi32>
        %1 = "tf.VarHandleOp"() {container = "", device = "/device:CPU:0", shared_name = "var_1"} : () -> tensor<!tf_type.resource<tensor<4xi32>>>
        "tf.AssignVariableOp"(%1, %cst_0) : (tensor<!tf_type.resource<tensor<4xi32>>>, tensor<4xi32>) -> ()

        return
      }
    }
  )mlir";

    // expected-remark@+1 {{Cost: 1}}
    %9 = "tf.StridedSlice"(%8, %5, %4, %4) {begin_mask = 0 : i64, device = "/job:localhost/replica:0/task:0/device:CPU:0", ellipsis_mask = 0 : i64, end_mask = 0 : i64, new_axis_mask = 0 : i64, shrink_axis_mask = 1 : i64} : (tensor<4xi32>, tensor<1xi32>, tensor<1xi32>, tensor<1xi32>) -> tensor<i32>
    // expected-remark@+1 {{Cost: 1}}

    %cst_1 = "tf.Const"() {value = dense<2> : tensor<4xi32>} : () -> tensor<4xi32>
    %var_1 = "tf.VarHandleOp"() {shared_name = "var_1"} : () -> tensor<!tf_type.resource<tensor<4xi32>>>
    "tf.AssignVariableOp"(%var_1, %cst_1) : (tensor<!tf_type.resource<tensor<4xi32>>>, tensor<4xi32>) -> ()
    return
  }

// CHECK: func.func @init_func_restore_op_multiple_variables

}

// CHECK-LABEL: conv2d_backprop_input_with_add
func.func @conv2d_backprop_input_with_add(%arg0: tensor<4xi32>, %arg1: tensor<3x3x1x32xf32>, %arg2: tensor<15x14x14x32xf32>) -> tensor<15x28x28x1xf32> {
  %0 = "tf.Conv2DBackpropInput"(%arg0, %arg1, %arg2) {strides = [1, 2, 2, 1], padding="SAME", dilations=[1, 1, 1, 1]}: (tensor<4xi32>, tensor<3x3x1x32xf32>, tensor<15x14x14x32xf32>) -> tensor<15x28x28x1xf32>

      %outputs_23, %control_24 = tf_executor.island(%control_18) wraps "tf.Const"() {_tpu_replicate = "cluster_sample_sequence", device = "", value = dense<[3, 32, 32, 1]> : tensor<4xi32>} : () -> tensor<4xi32>
      %outputs_25, %control_26 = tf_executor.island(%control_18) wraps "tf.TPUCompilationResult"() {_tpu_compilation_status = "cluster_sample_sequence", device = ""} : () -> tensor<!tf_type.string>

// CHECK-LABEL: broadcast_to_to_reshape
func.func @broadcast_to_to_reshape(%arg0: tensor<4x4x4xf32>, %arg1 : tensor<4xi32>) -> tensor<1x4x4x4xf32> {
  %0 = "tfl.broadcast_to"(%arg0, %arg1) : (tensor<4x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x4xf32>
  // CHECK: "tfl.reshape"
  // CHECK-SAME: (tensor<4x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x4xf32>
  func.return %0 : tensor<1x4x4x4xf32>
}

// CHECK-NEXT:      %14 = "tfl.pseudo_const"() <{value = dense<[-1, -1, 0, -1]> : tensor<4xi32>}> : () -> tensor<4xi32>
// CHECK-NEXT:      %15 = "tfl.pseudo_const"() <{value = dense<[-1, -1, -1, 0]> : tensor<4xi32>}> : () -> tensor<4xi32>
// CHECK-NEXT:      %16 = "tfl.pseudo_const"() <{value = dense<1> : tensor<i32>}> : () -> tensor<i32>

func.func @init() attributes {tf_saved_model.exported_names = ["__tf_saved_model_session_initializer_init"]} {
  %keys = "tf.Const"() {value = dense<[1, 2, 3, 4]> : tensor<4xi32> } : () -> tensor<4xi32>
  %values = "tf.Const"() {value = dense<[1, 2, 3, 4]> : tensor<4xi32> } : () -> tensor<4xi32>
  %handle = "tf.HashTableV2"() {container = "", key_dtype = i32, shared_name = "hash_table", use_node_name_sharing = false, value_dtype = i32} : () -> tensor<!tf_type.resource>