ASSERT_TRUE(out != nullptr);
  EXPECT_EQ(TF_FLOAT, TF_TensorType(out));
  EXPECT_EQ(2, TF_NumDims(out));
  EXPECT_EQ(4, TF_Dim(out, 0));
  EXPECT_EQ(1, TF_Dim(out, 1));
  float* values = static_cast<float*>(TF_TensorData(out));
  // These values are defined to be (input / 2) + 2.
  EXPECT_EQ(2, values[0]);
  EXPECT_EQ(2.5, values[1]);
  EXPECT_EQ(3, values[2]);
  EXPECT_EQ(3.5, values[3]);

    int64_t* dims = output_shapes->items[0].dims;

    if (!expected_shape.has_value()) {
      EXPECT_EQ(num_dims, -1);
      EXPECT_EQ(dims, nullptr);
      return;
    }

    EXPECT_EQ(num_dims, expected_shape->size());
    for (size_t i = 0; i < num_dims; ++i) {
      EXPECT_EQ(dims[i], (*expected_shape)[i]);
    }
    TF_DeleteShapeAndTypeList(input_shapes);

  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  float product[4] = {0};
  EXPECT_EQ(sizeof(product), TF_TensorByteSize(t));
  memcpy(&product[0], TF_TensorData(t), TF_TensorByteSize(t));
  TF_DeleteTensor(t);
  EXPECT_EQ(7, product[0]);
  EXPECT_EQ(10, product[1]);
  EXPECT_EQ(15, product[2]);
  EXPECT_EQ(22, product[3]);
  TF_DeleteStatus(status);
}

  std::unique_ptr<float[]> actual_values(new float[expected_values.size()]);
  EXPECT_EQ(sizeof(float) * expected_values.size(), TF_TensorByteSize(t));
  memcpy(actual_values.get(), TF_TensorData(t), TF_TensorByteSize(t));
  TF_DeleteTensor(t);

  for (int i = 0; i < expected_values.size(); i++) {
    EXPECT_EQ(expected_values[i], actual_values[i])
        << "Mismatch in expected values at (zero-based) index " << i;
  }

  TFE_TensorHandle* h = TestMatrixTensorHandle(ctx);
  EXPECT_EQ(TF_FLOAT, TFE_TensorHandleDataType(h));

  TF_Tensor* t = TFE_TensorHandleResolve(h, status.get());
  ASSERT_EQ(16, TF_TensorByteSize(t));
  float data[4] = {0};
  memcpy(&data[0], TF_TensorData(t), TF_TensorByteSize(t));
  EXPECT_EQ(1.0, data[0]);
  EXPECT_EQ(2.0, data[1]);
  EXPECT_EQ(3.0, data[2]);
  EXPECT_EQ(4.0, data[3]);
  TF_DeleteTensor(t);

  const DLTensor* dltensor_out = &dlm_out->dl_tensor;
  EXPECT_EQ(dltensor_out->device.device_type, dltensor_in->device.device_type);
  EXPECT_EQ(dltensor_out->device.device_id, dltensor_in->device.device_id);
  EXPECT_EQ(dltensor_out->ndim, dltensor_in->ndim);
  EXPECT_EQ(dltensor_out->dtype.code, dltensor_in->dtype.code);
  EXPECT_EQ(dltensor_out->dtype.bits, dltensor_in->dtype.bits);
  EXPECT_EQ(dltensor_out->dtype.lanes, dltensor_in->dtype.lanes);

  ASSERT_EQ(func_->record->fdef().signature().attr().size(), 2);
  EXPECT_EQ(func_->record->fdef().signature().attr(0).name(), "v1");
  EXPECT_EQ(func_->record->fdef().signature().attr(0).type(), "int");
  EXPECT_EQ(func_->record->fdef().signature().attr(1).name(), "v2");
  EXPECT_EQ(func_->record->fdef().signature().attr(1).type(), "int");
}

  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  EXPECT_EQ(TFE_TensorHandleDataType(packed_handle), TF_RESOURCE);
  EXPECT_EQ(TFE_TensorHandleNumDims(packed_handle, status), 0);
  EXPECT_EQ(TFE_TensorHandleNumElements(packed_handle, status), 1);

  const string composite_device_name =
      "/job:localhost/replica:0/task:0/device:COMPOSITE:0";
  EXPECT_EQ(TFE_TensorHandleDeviceName(packed_handle, status),

  const std::vector<std::unique_ptr<ParallelTensor>>& handles = *outputs;
  const std::vector<int64_t>* shape;
  absl::Status s = handles[0]->Shape(&shape);
  ASSERT_TRUE(s.ok());
  EXPECT_EQ(0, shape->size());
}

TEST(PARALLEL_DEVICE_LIB, TestCancelOnError) {
  std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)> status(
      TF_NewStatus(), TF_DeleteStatus);

                  .ok());
  ASSERT_TRUE(worker_server2->Start().ok());

  TFE_ContextSetServerDef(context.get(), 0, serialized.data(),
                          serialized.size(), status.get());
  EXPECT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  BasicTestsForTwoDevices(context.get(),
                          "/job:worker/replica:0/task:1/device:CPU:0",