%4 = "tfl.minimum"(%0, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %5 = "tfl.minimum"(%1, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %6 = "tfl.minimum"(%2, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %7 = "tfl.minimum"(%3, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>

// CHECK-SAME: (%[[ARG_0:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_1:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_2:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_3:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_4:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>)

      %0 = tf_executor.graph {
        %read0, %read0_control = tf_executor.island wraps "tf.ReadVariableOp"(%arg0) : (tensor<*x!tf_type.resource<tensor<32xf32>>>) -> tensor<32xf32>
        %assign0_control = tf_executor.island(%read0_control) wraps "tf.AssignVariableOp"(%arg0, %arg1) : (tensor<*x!tf_type.resource<tensor<32xf32>>>, tensor<32xf32>) -> ()

  // Use other output
  %3:6 = "tf.FusedBatchNormV3"( %2#0, %arg1, %arg2, %arg3, %arg4) {T = "tfdtype$DT_FLOAT", U = "tfdtype$DT_FLOAT", data_format = "NHWC", epsilon = 0.001 : f32, is_training = false} : (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>) -> (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>)

    return %2 : tensor<1x2xf32>
  }

  func.func private @composite_add_fn(%arg0: tensor<1x2xf32>, %arg1: tensor<1x2xf32>) -> tensor<1x2xf32> attributes {_from_xla_call_module} {
    %0 = stablehlo.add %arg0, %arg1 : tensor<1x2xf32>
    %1 = stablehlo.add %0, %arg1 : tensor<1x2xf32>
    return %1 : tensor<1x2xf32>
  }
}

// -----

  func.return %1 : tensor<3xf32>
}

// CHECK-LABEL: func @squared_difference_real
func.func @squared_difference_real(%arg0: tensor<3xf32>, %arg1: tensor<3xf32>) -> tensor<3xf32> {
  // CHECK: [[R1:%.+]] = "tf.Sub"(%arg0, %arg1)
  // CHECK: "tf.Mul"([[R1]], [[R1]])
  %1 = "tf.SquaredDifference"(%arg0, %arg1) : (tensor<3xf32>, tensor<3xf32>) -> tensor<3xf32>
  func.return %1 : tensor<3xf32>
}

    %14 = "tf.Cast"(%13) {Truncate = false, device = ""} : (tensor<?x?x?x?x2xi32>) -> tensor<?x?x?x?x2xf32>
    %15 = "tf.Mul"(%14, %cst) {device = ""} : (tensor<?x?x?x?x2xf32>, tensor<2xf32>) -> tensor<?x?x?x?x2xf32>
    %16 = "tf.Identity"(%15) {device = ""} : (tensor<?x?x?x?x2xf32>) -> tensor<?x?x?x?x2xf32>
    return %16 : tensor<?x?x?x?x2xf32>
  }

  // CHECK-SAME: (tensor<5x3xf32>, tensor<2xi32>, tensor<2xi32>) -> tensor<1x3xf32>
  // CHECK: %[[UPDATE_SLICE0_START:.*]] = "tf.Const"() <{value = dense<0> : tensor<2xi32>}> : () -> tensor<2xi32>
  // CHECK: %[[UPDATE_SLICE0:.*]] = "tf.Slice"(%[[GATHER]], %[[UPDATE_SLICE0_START]], %[[SLICE_SIZE]]) : (tensor<2x3xf32>, tensor<2xi32>, tensor<2xi32>) -> tensor<1x3xf32>

        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<4x2xf32>, tensor<4xf32>,
        tensor<1x4xf32>, tensor<1x2xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>) -> tensor<*xf32>
  %24 = "quantfork.stats"(%23) {layerStats = dense<[-1.0, 2.0]> : tensor<2xf32>} : (tensor<*xf32>) -> tensor<*xf32>

    %0 = "tf.MatMul"(%arg0, %arg1) {attr_map = "0:transpose_a,1:transpose_b", device = "", transpose_a = false, transpose_b = false} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
    return %0 : tensor<2x2xf32>
  }