// 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>)

// CHECK-LABEL: func @init
func.func @init(%arg0: tensor<4xf32>) -> tensor<7xf32> {
    %0 = builtin.unrealized_conversion_cast to tensor<7xf32>
    return %0 : tensor<7xf32>
}

// CHECK-LABEL: func @next
func.func @next(%arg0: tensor<7xf32>, %arg1: tensor<3xf32>) -> tensor<6xf32> {
    %0 = builtin.unrealized_conversion_cast to tensor<6xf32>
    return %0 : tensor<6xf32>
}

// CHECK-LABEL: func @finalize

    // CHECK-DAG: %[[SIGN:.*]] = "tf.Sign"(%[[PROX]]) : (tensor<4xf32>) -> tensor<4xf32>
    // CHECK-DAG: %[[ABS:.*]] = "tf.Abs"(%[[PROX]]) : (tensor<4xf32>) -> tensor<4xf32>
    // CHECK-DAG: %[[SCALED_L1:.*]] = "tf.Mul"(%[[ADAGRAD_LR]], %[[L1]]) : (tensor<4xf32>, tensor<f32>) -> tensor<4xf32>
    // CHECK-DAG: %[[PROX_NEW:.*]] = "tf.Sub"(%[[ABS]], %[[SCALED_L1]]) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>

func.func @tensorlistGetItemWithUnknownRank(%arg0: tensor<*xf32>, %arg1: tensor<1xi32>, %arg2: tensor<i32>) -> (tensor<*xf32>, tensor<*xf32>) {
  %0 = "tf.TensorListFromTensor"(%arg0, %arg1) : (tensor<*xf32>, tensor<1xi32>) -> tensor<!tf_type.variant<tensor<*xf32>>>
  %1 = "tf.TensorListGetItem"(%0, %arg2, %arg1) : (tensor<!tf_type.variant<tensor<*xf32>>>, tensor<i32>, tensor<1xi32>) -> tensor<*xf32>

  %cst_zero = arith.constant dense<0.0> : tensor<f32>
  %cst_one = arith.constant dense<1.0> : tensor<f32>

  // CHECK-DAG: %[[RES_NO_FOLD:.*]] = "tf.Pow"(%arg0, %arg1)
  %0 = "tf.Pow"(%arg0, %arg1) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>

  // CHECK-DAG: %[[POW_ZERO:.*]] = "tf.Const"() <{value = dense<1.000000e+00> : tensor<4xf32>}> : () -> tensor<4xf32>

  %1 = "tf.Mul"(%arg1, %arg1) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  %2 = "tf.Div"(%arg1, %arg1) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>

  // CHECK: "tf.If"{{.+}}else_branch = @tf.IfRegion_else{{.+}}then_branch = @tf.IfRegion_then
  %3 = "tf.IfRegion"(%arg0) ({
     %4 = "tf.Select"(%arg0, %0, %1) : (tensor<i1>, tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
     "tf.Yield"(%4) : (tensor<2xf32>) -> ()

    %1 = "tf.BiasAdd"(%0, %cst_0) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<3xf32>) -> tensor<*xf32>
    %2 = "tf.PartitionedCall"(%arg0, %cst_2) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_depthwise_conv2d_fn_1} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>
    func.return %1, %2: tensor<*xf32>, tensor<*xf32>

  %result = tf_executor.graph {

// CHECK: tf_executor.Merge{{.*}}(tensor<*xf32>, tensor<4xf32>) -> (tensor<4xf32>, tensor<i32>, !tf_executor.control)
    %value, %idx, %ctlMerge = tf_executor.Merge %arg0, %arg1  : (tensor<*xf32>, tensor<4xf32>) -> (tensor<4xf32>, tensor<i32>, !tf_executor.control)
    tf_executor.fetch %value : tensor<4xf32>
  }

  %0 = "quantfork.stats"(%arg0) {
    layerStats = dense<[-1.28e-5, 1.27e-5]> : tensor<2xf32>
  } : (tensor<1x5x5x2xf32>) -> tensor<1x5x5x2xf32>
  %w = arith.constant dense<[[[[-1.0, 1.0]]], [[[1.0, 2.0]]], [[[-2.0, 1.0]]]]> : tensor<3x1x1x2xf32>
  %b = arith.constant dense<0.0> : tensor<3xf32>
  %b2 = arith.constant dense<[1.0e-2, 2.1473647e1, -2.1473647e2]> : tensor<3xf32>
  %conv = "tfl.conv_2d"(%0, %w, %b) {

  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "ijk,j->i"}: (tensor<4x5x6xf32>, tensor<5xf32>) -> tensor<4xf32>
  func.return %0 : tensor<4xf32>
// CHECK-LABEL: einsum_no_match
// CHECK: %[[v0:.*]] = "tf.Einsum"(%arg0, %arg1) <{equation = "ijk,j->i"}> {T = "tfdtype$DT_FLOAT"} : (tensor<4x5x6xf32>, tensor<5xf32>) -> tensor<4xf32>
// CHECK: return %[[v0]]
}