#include "tensorflow/core/protobuf/tensorflow_server.pb.h"

namespace {

using ::tensorflow::string;

void ReplaceTaskInServerDef(tensorflow::ServerDef* server_def, int task_index) {
  tensorflow::JobDef* job_def = server_def->mutable_cluster()->mutable_job(0);
  int port = tensorflow::testing::PickUnusedPortOrDie();
  job_def->mutable_tasks()->at(task_index) =
      tensorflow::strings::StrCat("localhost:", port);
}

                         {12, 15, 18, 12, 15, 18, 12, 15, 18},
                         {6, 6, 6, 15, 15, 15, 24, 24, 24},
                         {6, 15, 24, 6, 15, 24, 6, 15, 24}};

  for (int i{}; i < 4; ++i) {
    bool transpose_a = transpose_a_vals[i];
    bool transpose_b = transpose_b_vals[i];
    Model MatMulModel =
        [transpose_a, transpose_b](
            AbstractContext* ctx,

    const TF_Graph* fn_body, const char* fn_name, int ninputs,
    const TF_Output* inputs, std::vector<OutputTensor>* input_tensors,
    std::unordered_map<const Node*, std::vector<int>>* input_nodes)
    TF_EXCLUSIVE_LOCKS_REQUIRED(fn_body->mu) {
  input_tensors->reserve(ninputs);
  for (int i = 0; i < ninputs; ++i) {
    Node* node = inputs[i].oper ? &inputs[i].oper->node : nullptr;
    int idx = inputs[i].index;

  }
  TFE_OpAddAttrs(op_.get(), attributes);
  for (int input_index = 0; input_index < inputs.size(); ++input_index) {
    TFE_OpAddInput(op_.get(), inputs[input_index], status);
    if (TF_GetCode(status) != TF_OK) return;
  }
  std::vector<TFE_TensorHandle*> unwrapped_results(expected_max_outputs);
  int real_num_outputs = expected_max_outputs;

  TFE_OpSetDevice(op.get(), device, status);
  if (TF_GetCode(status) != TF_OK) return;

  TFE_TensorHandle* result_handles[num_replicas];
  int num_retvals = num_replicas;
  TFE_Execute(op.get(), result_handles, &num_retvals, status);
  if (TF_GetCode(status) != TF_OK) return;
  for (int i = 0; i < num_replicas; ++i) {
    (*components)[i].reset(result_handles[i]);
  }
}

      reinterpret_cast<NamedParallelDevice*>(device_info);
  std::vector<MaybeParallelTensorUnowned> typed_inputs;
  int num_inputs = TFE_OpGetFlatInputCount(original_op, status);
  if (TF_GetCode(status) != TF_OK) return;
  typed_inputs.reserve(num_inputs);
  for (int i = 0; i < num_inputs; ++i) {
    TFE_TensorHandle* input = TFE_OpGetFlatInput(original_op, i, status);
    if (TF_GetCode(status) != TF_OK) return;

 public:
  // Eager async execution is only supported when remote eager is not in use
  // (b/157523095).
  explicit ParallelDevice(const std::vector<std::string>& devices,
                          bool is_async = false, int in_flight_nodes_limit = 0);

  ~ParallelDevice();

  // Helper to copy a tensor handle from another device once for each component
  // of the ParallelDevice.
  //

  // by the underlying tensorflow::VariableInfo.
  TF_VariableInfo(int index, const std::string& name, tensorflow::Var* var) {
    var_info = tensorflow::VariableInfo{index, name, var};
  }

  tensorflow::VariableInfo var_info{0, "", nullptr};
};

TF_VariableInfo* TF_CreateVariableInfoFromContext(TF_OpKernelContext* ctx,
                                                  int index,

    TFE_TensorHandle* result_handles;
    int num_retvals = 1;
    TFE_Execute(op.get(), &result_handles, &num_retvals, status.get());
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_INVALID_ARGUMENT)
        << TF_Message(status.get());
  }
}

TensorHandlePtr CollectiveSum(TFE_Context* context, TFE_TensorHandle* input,
                              int group_size, TF_Status* status) {

  // When remote is false, function device is placed on task0. Handle types are
  // REMOTE, REMOTE, LOCAL on task0. When remote is true, function device is
  // placed on task1, Handle types are LOCAL, REMOTE, LOCAL on task1.
  int num_replicas = 3;
  std::vector<TFE_TensorHandle*> handles = {h0, h1, h2};
  TFE_TensorHandle* packed_handle =
      TFE_CreatePackedTensorHandle(ctx, handles.data(), &num_replicas, status);