bool heavy_load_on_streaming_rpc,
                                   bool remote_func_outputs = false,
                                   bool has_packed_input = false);

    ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
    ASSERT_TRUE(absl::StrContains(device_name, "GPU:0")) << device_name;

    // .backing_device of shape is CPU since the tensor is backed by CPU
    backing_device_name =
        TFE_TensorHandleBackingDeviceName(retvals[0], status.get());
    ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

}

void NoOpDeallocator(void* data, size_t, void*) {}

TEST(CAPI, MalformedTensor) {
  // See https://github.com/tensorflow/tensorflow/issues/7394
  // num_dims = 0 implies a scalar, so should be backed by at least 4 bytes of
  // data.
  TF_Tensor* t =
      TF_NewTensor(TF_FLOAT, nullptr, 0, nullptr, 0, &NoOpDeallocator, nullptr);
  ASSERT_TRUE(t == nullptr);
}

TEST(CAPI, AllocateTensor) {

  TFE_TensorHandle* packed_handle =
      TFE_CreatePackedTensorHandle(ctx, handles.data(), &num_replicas, status);
  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);

  if (*requested_placement == '\0') {
    TF_SetStatus(
        status, TF_INTERNAL,
        "Ops must be placed on the parallel device explicitly, or their inputs "
        "first un-packed. Got an un-placed op with an input placed on the "
        "parallel device.");
    return;
  }
  TFE_Context* context = TFE_OpGetContext(original_op, status);
  if (TF_GetCode(status) != TF_OK) return;

                                       bool remote_func_outputs = false,
                                       bool has_packed_input = false) {
  return TestRemoteExecuteSilentCopies(async, remote, /*func=*/true,
                                       heavy_load_on_streaming_rpc,
                                       remote_func_outputs, has_packed_input);
}

TEST(CAPI, RemoteExecuteSilentCopiesAsyncFunc) {

    // While the Accumulate method is running (accumulating is True), any op
    // executions not forwarded to backward_tape should be ignored.
    bool accumulating;
  };
  // A deque-backed stack, whose element references are not invalidated by
  // pushes and pops at the back.
  std::stack<AccumulatorCallState> call_state_;
};

// Template instantiations here

TF_CAPI_EXPORT TFE_TensorHandle* TFE_NewTensorHandleFromTensor(
    TFE_Context* ctx, TF_Tensor* t, TF_Status* status);

// Create a packed TensorHandle with the given list of TensorHandles.
// If `handles` are on the same device, assign the same device to the packed
// handle; if `handles` are on different deivces, assign a CompositeDevice to
// it.
TF_CAPI_EXPORT extern TFE_TensorHandle* TFE_CreatePackedTensorHandle(

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

  TFE_TensorHandle* packed_handle = nullptr;
  if (has_packed_input) {
    int num_replicas = 1;
    std::vector<TFE_TensorHandle*> packed_handles = {h1_task2};
    packed_handle = TFE_CreatePackedTensorHandle(ctx, packed_handles.data(),
                                                 &num_replicas, status);

      Status* status) = 0;

  // Create an operation to perform op execution
  ImmediateExecutionOperation* CreateOperation() override = 0;

  // Returns whether the runtime is backed by TFRT or the legacy TF Eager
  // Runtime. This is necessary to decouple runtime-dependent
  // code that is layered on top of the runtime.
  virtual bool UsesTFRT() = 0;

  // List attributes of available devices