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]);

  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;
  }

  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",

  TFE_Context* ctx = TFE_NewContext(opts, status);
  EXPECT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  TFE_ContextSetServerDefWithTimeout(ctx, 0, serialized_server_def.data(),
                                     serialized_server_def.size(),
                                     init_timeout_in_ms, status,
                                     /*clear_existing_contexts=*/false);
  EXPECT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);

  // Copying on to a parallel device must be explicit.
  TensorHandlePtr failed_copy_on_result(TFE_TensorHandleCopyToDevice(
      cpu_value.get(), context.get(), device_name, status.get()));
  EXPECT_EQ(TF_GetCode(status.get()), TF_UNIMPLEMENTED);

  std::array<TFE_TensorHandle*, 2> components{cpu_value.get(), cpu_value.get()};
  TensorHandlePtr device_value = CreatePerDeviceValues(

  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  TF_Tensor* result_tensor = TFE_TensorHandleResolve(result_t, status.get());
  float* result_value = static_cast<float*>(TF_TensorData(result_tensor));
  EXPECT_EQ(*result_value, 4.0);

  TF_DeleteTensor(result_tensor);
  TF_DeleteAbstractTensor(result);
  TF_DeleteOutputList(o);
  TF_DeleteExecutionContext(ctx);
}

// MatMul Test