if (!device_name.empty()) {
    TFE_OpSetDevice(op, device_name.c_str(), status);
  }
  if (TF_GetCode(status) != TF_OK) return nullptr;
  TFE_TensorHandle* var_handle = nullptr;
  int num_retvals = 1;
  TFE_Execute(op, &var_handle, &num_retvals, status);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  TFE_DeleteOp(op);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  CHECK_EQ(1, num_retvals);

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

  TFE_TensorHandle* result_handle;
  int num_retvals = 1;
  TFE_Execute(op.get(), &result_handle, &num_retvals, status);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  return TensorHandlePtr(result_handle);
}

template <typename value_type>

      TFE_OpAddInput(op, t->tensor, s);
    } else {
      TFE_OpAddInput(op, input, s);
    }
    if (TF_GetCode(s) != TF_OK) return;
  }
  std::vector<TFE_TensorHandle*> op_outputs(*num_outputs);
  TFE_Execute(op, op_outputs.data(), num_outputs, s);
  TFE_DeleteOp(op);
  if (TF_GetCode(s) != TF_OK) return;
  std::vector<TFE_TensorHandle*> unwrapped_outputs;
  unwrapped_outputs.reserve(op_outputs.size());

                    absl::optional<int64_t> step_id = absl::nullopt);
  // Block until the previous `StartExecute` operation has executed. Forwards
  // the status from `TFE_Execute` and returns outputs if the status is OK.
  std::vector<TensorHandlePtr> Join(TF_Status* status);

  // Block until all Ops finished running on the thread.
  void AsyncWait(TF_Status* status);

 private:

    // The input handles should never change since they have been mirrored.
    ASSERT_FALSE(remote_arg->HasLocalMirror(nullptr));
  }

  TFE_TensorHandle* retvals[1];
  int num_retvals = 1;
  TFE_Execute(matmul, &retvals[0], &num_retvals, status);
  EXPECT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);

  // TODO(gjn): Add support for waiting on async local mirrors
  if (!remote && !async && !remote_func_outputs) {

                                                TF_Status* status) {
  int ret = -1;
  status->status = tensorflow::unwrap(op)->OutputLength(output_name, &ret);
  return ret;
}

void TFE_Execute(TFE_Op* op, TFE_TensorHandle** retvals, int* num_retvals,
                 TF_Status* status) {
  tensorflow::ImmediateExecutionOperation* unwrapped_op =
      tensorflow::unwrap(op);

  status->status =