ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  TFE_OpSetAttrType(handle_op.get(), "dtype", TF_FLOAT);
  TFE_OpSetAttrShape(handle_op.get(), "shape", /*dims=*/nullptr, /*num_dims=*/0,
                     status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  auto outputs =

    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    TF_OutputListPushBack(func_outputs, add_output2, s);
    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    TF_OutputListPushBack(func_outputs, mm_output, s);
    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    func = TF_FinalizeFunction(graph_ctx, func_outputs, s);
    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    TF_DeleteOutputList(func_outputs);
  }

  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

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

  auto* retval_task0 =
      TFE_TensorHandleCopyToDevice(retvals[0], ctx, local_device_name, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* b = Placeholder(graph, s, "B");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* two = ScalarConst(2, graph, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* plus2 = Add(a, two, graph, s, "plus2");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  TFE_OpAddInput(func, packed_handle, status);
  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  if (remote) {
    TFE_OpSetDevice(func, task1_name, status);
    ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  }

  TFE_TensorHandle* retvals[1] = {nullptr};
  int num_retvals = 1;
  TFE_Execute(func, &retvals[0], &num_retvals, status);

      TF_SetStatus(first_bad_status.get(), TF_GetCode(async_wait_status.get()),
                   TF_Message(async_wait_status.get()));
    }
  }

  if (first_bad_status != nullptr) {
    TF_SetStatus(status, TF_GetCode(first_bad_status.get()),
                 TF_Message(first_bad_status.get()));
  }
}

absl::optional<std::vector<std::unique_ptr<ParallelTensor>>>
ParallelDevice::Join(