if (use_resource) {
      Output var =
          ops::VarHandleOp(scope.WithOpName("var"), DataType::DT_FLOAT, {});
      read_var = ops::ReadVariableOp(
          scope.WithOpName("var/Read/ReadVariableOp"), var, DataType::DT_FLOAT);
      auto assign = ops::AssignVariableOp(scope.WithOpName("assign"), var, a);
      Output var_1 =
          ops::VarHandleOp(scope.WithOpName("var_1"), DataType::DT_FLOAT, {});
      Output read_var_1 =

// GraphDef.
std::unique_ptr<OperationPass<ModuleOp>> CreateMergeSaveFunctionOpsToMainPass();

// Creates a pass that "unfreezes" ConstOps into variables. Each ConstOp's use
// will be replaced by a VarHandleOp -> ReadVariableOp pattern. The newly
// created variables will be initialized in the session initializer function via
// AssignVariableOps.
std::unique_ptr<OperationPass<ModuleOp>> CreateUnfreezeConstantsPass();

             %arg3: !tf_res_md_f32 {tf.device = "/device:TPU:1"}) {

    %0 = "tf.Const"() {value = dense<100> : tensor<i32>} : () -> tensor<i32>
    // CHECK: %[[STATE0:.*]] = "tf.VarHandleOp"()
    // CHECK-SAME: device = "/device:TPU:0"
    // CHECK: %[[STATE1:.*]] = "tf.VarHandleOp"()
    // CHECK-SAME: device = "/device:TPU:1"
    // CHECK: %[[WHILE:.*]] = "tf.WhileRegion"(
    %1 = "tf.WhileRegion"(%0) ({
       // Condition region

// CHECK: }

func.func @unknown_side_effecting_op(%arg0: tensor<32xf32>) -> () {
  tf_executor.graph {
    %island = tf_executor.island {
      %vh0 = "tf.VarHandleOp"() {container = "c", shared_name = "v0"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>
      %vh1 = "tf.VarHandleOp"() {container = "c", shared_name = "v1"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>

    resource variables directly..

    Here is a simple example in TensorFlow where a device doubles the value of a
    TensorFlow resource variable and returns new value:

    ```mlir
    %resource_handle = "tf.VarHandleOp"()
    %1 = "tf_device.cluster"() ( {
      %init_value = "tf.ReadVariableOp"(%resource_handle)
      "tf.AssignAddVariableOp"(%resource_handle, %init_value)
      %new_value = "tf.ReadVariableOp"(%resource_handle)

// CHECK: }

func.func @unknown_side_effecting_op(%arg0: tensor<32xf32>) -> () {
  tf_executor.graph {
    %island = tf_executor.island {
      %vh0 = "tf.VarHandleOp"() {container = "c", shared_name = "v0"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>
      %vh1 = "tf.VarHandleOp"() {container = "c", shared_name = "v1"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>

func.func @unknown_side_effecting_op(%arg0: tensor<32xf32>) {
  tf_executor.graph {
    %outputs, %control = tf_executor.island wraps "tf.VarHandleOp"() {container = "c", shared_name = "v0"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>
    %outputs_0, %control_1 = tf_executor.island wraps "tf.VarHandleOp"() {container = "c", shared_name = "v1"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>

                                                                               \
  REGISTER_KERNEL_BUILDER(                                                     \
      Name("VarHandleOp").Device(DEVICE).HostMemory("resource"), VarHandleOp); \
  REGISTER_KERNEL_BUILDER(                                                     \
      Name("_VarHandlesOp").Device(DEVICE).HostMemory("resources"),            \

// Creates the per-device variables that represent the formatting state of each
// device.
llvm::SmallVector<TF::VarHandleOp, 4> CreateStateVars(
    const llvm::SmallDenseMap<llvm::StringRef, llvm::SmallVector<StringRef, 4>>&
        devices,
    Location loc, RankedTensorType key_type, OpBuilder* builder) {
  llvm::SmallVector<TF::VarHandleOp, 4> state_vars;

  // TODO(b/148913020): Remove this constraint once model parallelism is
  // supported.

  Output var_handle =
      ops::VarHandleOp(scope.WithOpName("Var" + id), DT_FLOAT, TensorShape({}));
  Output read =
      ops::ReadVariableOp(scope.WithOpName("Read" + id), var_handle, DT_FLOAT);
  return read.node();
}

Node* MakeWrite(const Scope& scope, const string& id) {
  Output var_handle =
      ops::VarHandleOp(scope.WithOpName("Var" + id), DT_FLOAT, TensorShape({}));
  Output value_to_write =