std::vector<tensorflow::Output> outputs(n);
  for (int i = 0; i < n; ++i) {
    outputs[i] =
        tensorflow::Output(&tf_outputs[i].oper->node, tf_outputs[i].index);
  }
  return outputs;
}

void TFOutputsFromOutputs(const std::vector<tensorflow::Output>& outputs,
                          TF_Output* tf_outputs) {
  for (int i = 0; i < outputs.size(); i++) {

  NodeBuilder node_builder(opts.WithName(name).GetNameForOp("_XlaRecvAtHost"),
                           "_XlaRecvAtHost", opts.op_registry());
  node_builder.Input(std::move(key_input));
  return opts.WithAttr("Toutputs", dtypes)
      .WithAttr("key", key)
      .WithAttr("device_ordinal", 0)
      .WithAttr("_encapsulate", cluster)
      .WithAttr("_outside", oc_cluster)
      .FinalizeBuilder(&node_builder);
}

                void transform(TransformOutputs outputs) {
                    def input = inputArtifact.get().asFile
                    File outputA = outputs.file(input.name + ".A.txt")
                    assert outputA.parentFile.directory && outputA.parentFile.list().length == 0
                    outputA.text = "Output A"

                    File outputB = outputs.file(input.name + ".B.txt")

                    inputArtifact.get().asFile
                }

                void transform(TransformOutputs outputs) {
                    def output = outputs.file(input.name + ".txt")
                    println "Transformed \$input.name to \$input.name into \${output.parentFile}"
                    outputs.file(inputArtifact)
                    output.text = String.valueOf(input.length())
                }
            }

	var (
		gp01           = regInfo{inputs: nil, outputs: gponly}
		gp11           = regInfo{inputs: []regMask{gp}, outputs: gponly}
		gp11sp         = regInfo{inputs: []regMask{gpsp}, outputs: gponly}
		gp11sb         = regInfo{inputs: []regMask{gpspsbg}, outputs: gponly}
		gp21           = regInfo{inputs: []regMask{gp, gp}, outputs: gponly}
		gp21sp         = regInfo{inputs: []regMask{gpsp, gp}, outputs: gponly}

  std::vector<TensorId> outputs;
  output_node_names.reserve(specs.outputs.size());
  for (const auto& output : specs.outputs) {
    TensorId tensor = ParseTensorName(output);
    auto remapped_it = remapped_feeds_.find(tensor);
    if (remapped_it != remapped_feeds_.end()) {
      output_node_names.insert(remapped_it->second);
      outputs.push_back({remapped_it->second, 0});
    } else {

  for (const auto& signature_def_data : signature_defs) {
    auto inputs = GetList(subgraph_index, signature_def_data.inputs);
    auto outputs = GetList(subgraph_index, signature_def_data.outputs);
    auto inputs_buf = builder_.CreateVector(inputs);
    auto outputs_buf = builder_.CreateVector(outputs);
    auto signature_key_buf =
        builder_.CreateString(signature_def_data.signature_key);

}

def ExecutorConvertControlToDataOutputsPass : Pass<"tf-executor-convert-control-to-data-outputs", "ModuleOp"> {
  let summary = "Chain control outputs of while loop body";

  let description = [{
    This pass converts the control outputs of a while loop body function to data
    outputs. Thus, inter iteration control dependencies are transformed to

    graphdef = loader.get_meta_graph_def_from_tags(tags).graph_def
    self.assertTrue(self._contains_op(graphdef, 'XlaDotV2'))

    outputs = converted_model.signatures[signature_key](**model_inputs)

    self.assertAllClose(outputs, expected_outputs, atol=1e-1)

  # Equations NOT supported for XLA operations.
  @parameterized.parameters(
      testing.parameter_combinations([{

    func.return %4, %3 : tensor<*xi32>, tensor<*xi1>
  }
}

// -----

// Tests that outputs are correctly merged and fed from TPU computation for
// tiled output sharding.

// The following OpSharding is used for TPU computation outputs in below test:
// Proto debug string:
//  output 0
//   type: OTHER
//   tile_assignment_dimensions: 1