// CHECK: %[[CONST1:.*]] = "tf.Const"() <{value = dense<1> : tensor<i32>}> : () -> tensor<i32>
  %const1 = "tf.Const"() {value = dense<1> : tensor<i32>} : () -> tensor<i32>
  // CHECK: %[[SUB:.*]] = "tf.Sub"(%[[BARG1]], %[[CONST1]])
  %sub = "tf.Sub"(%arg1, %const1) : (tensor<i32>, tensor<i32>) -> tensor<i32>
  %elem = "tf._SomeOp"() : () -> tensor<3xf32>

  Output const0 = ops::Const(root.WithOpName("const0"), 1);
  Output const1 = ops::Const(root.WithOpName("const1"), 2);

  Output add = ops::Add(root.WithOpName("add"), const0, const1);

  root.graph()->AddControlEdge(id0.node(), const0.node());
  root.graph()->AddControlEdge(id1.node(), const1.node());

  std::unique_ptr<DeadnessAnalysis> result;

  // done in parallel.
  //
  // This graph is:
  // (Const0, Const0) -> MatMul0
  // (Const1, Const1) -> MatMul1
  // (MatMul0, MatMul1) -> MatMulCombined
  //
  // Device0: [Const0, Const0, MatMul0]
  // Device1: [Const1, Const1, MatMul1, MatMulCombined]
  //
  // Cluster0: [Const0, Const0, MatMul0]
  // Cluster1: [Const1, Const1, MatMul1]
  // Cluster2: [MatMulCombined]

    //        |         |
    //      Const_0    Const_1
    //
    const float const0_val[] = {1.0, 2.0, 3.0, 4.0};
    const float const1_val[] = {1.0, 0.0, 0.0, 1.0};
    TF_Operation* const0 = FloatConst2x2(graph_, s_, const0_val, "Const_0");
    TF_Operation* const1 = FloatConst2x2(graph_, s_, const1_val, "Const_1");
    TF_Operation* matmul = MatMul(graph_, s_, const0, const1, "MatMul");
    inputs[0] = TF_Output{const0, 0};

    -> (tensor<*x!tf_type.resource<tensor<4xf32>>>, tensor<4xf32>) {
  // CHECK-NEXT: %[[CONST:.*]] = "tf.Const"()
  %constant = "tf.Const"() {value = dense<0.0> : tensor<4xf32>} : () -> tensor<4xf32>
  "tf.AssignVariableOp"(%arg0, %constant) : (tensor<*x!tf_type.resource<tensor<4xf32>>>, tensor<4xf32>) -> ()
  // CHECK-NEXT: return %[[CONST]], %[[CONST]]
  func.return %arg0, %constant : tensor<*x!tf_type.resource<tensor<4xf32>>>, tensor<4xf32>
}

    tf_device.return %identity : tensor<i32>
  }) : () -> (tensor<i32>)
  return %cluster : tensor<i32>
}
```

will be transformed into:

```mlir
func @cluster() -> tensor<i32> {
  %cluster = "tf_device.cluster"() ( {
    %const = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>
    %identity = "tf.Identity"(%const) : (tensor<i32>) -> tensor<i32>

  %3 = stablehlo.add %2, %1 : tensor<1x3xf32>
  func.return %3: tensor<1x3xf32>
}
// CHECK: %[[CONST_0:.*]] = stablehlo.constant dense<2.000000e+00>
// CHECK: %[[CONST_1:.*]] = stablehlo.constant dense<2.000000e+00>
// CHECK: %[[XLA_CALL_MODULE:.*]] = "tf.XlaCallModule"(%arg0, %[[CONST_0]], %[[CONST_1]])
// CHECK: return %[[XLA_CALL_MODULE:.*]] : tensor<1x3xf32>
// CHECK: }

	for {
		x := v_0
		s := v_1
		v.reset(OpInt64Make)

// CHECK-DAG: %[[CONST_8:.*]] = "tf.Const"() <{value = dense<[162, 160]> : tensor<2xi32>}> : () -> tensor<2xi32>
// CHECK: %[[PADV2_0:.*]] = "tf.PadV2"({{.*}}, %[[CONST_4]], %[[CONST_5]]) : (tensor<1x3x4x3xi8>, tensor<4x2xi32>, tensor<i8>) -> tensor<1x4x5x3xi8>
// CHECK: %[[XLACONVV2_0:.*]] = "tf.XlaConvV2"(%[[PADV2_0]], %[[CONST_6]], %[[CONST_0]], %[[CONST_3]], %[[CONST_1]], %[[CONST_1]], %[[CONST_2]])

// CHECK-DAG: %[[CONST_2:.+]] = stablehlo.constant dense<3> : tensor<1xi32>
// CHECK-DAG: %[[CONST_3:.+]] = stablehlo.constant dense<4> : tensor<1xi32>
// CHECK-DAG: %[[CONST_4:.+]] = stablehlo.constant dense<2> : tensor<1xi32>