TEST(DeadnessAnalysisTest, Recv) {
  Scope root = Scope::NewRootScope().ExitOnError();

  Output recv_a = ops::_Recv(root.WithOpName("recv_a"), DT_FLOAT, "tensor_a",
                             "sender", 0, "receiver");
  Output recv_b = ops::_Recv(root.WithOpName("recv_b"), DT_FLOAT, "tensor_b",
                             "sender", 0, "receiver");
  Output add = ops::Add(root.WithOpName("add"), recv_a, recv_b);

TEST(XlaCompilationTest, ClusterControlTrigger) {
  Scope root = Scope::NewRootScope().ExitOnError();

  Output recv_a = ops::_Recv(root.WithOpName("recv_a"), DT_BOOL, "tensor_a",
                             "sender", 0, "receiver");
  Output recv_b = ops::_Recv(root.WithOpName("recv_b"), DT_BOOL, "tensor_b",
                             "sender", 0, "receiver");
  Output const_a = ops::Const(root.WithOpName("const_a"), 42);

    Gradient* grad_tensor = grad[grad_index];
    if (grad_tensor != nullptr) {
      int64_t tensor_id = vspace_.TensorId(grad_tensor);
      targets.push_back(tensor_id);
      if (sources_set.find(tensor_id) != sources_set.end()) {
        sources_that_are_targets.emplace(
            tensor_id, vspace_.TapeTensorFromGradient(grad_tensor));
      }
      Gradient* in_grad = in_grads[grad_index];

  // "tensor_id() is live and evaluates to true".
  //
  // If `must_be_true()` is false then this SymbolPredicate represents the
  // proposition "tensor_id() is live (and may evaluate to any value)"
  TensorId tensor_id() const { return tensor_id_; }
  bool must_be_true() const { return must_be_true_; }

 private:
  TensorId tensor_id_;
  bool must_be_true_;
};

// Tensor types
def TF_BoolTensor : TensorOf<[TF_Bool]>;

def TF_IntTensor : TensorOf<[TF_Int]>;
def TF_Int4Tensor : TensorOf<[TF_Int4]>;
def TF_Int8Tensor : TensorOf<[TF_Int8]>;
def TF_Int16Tensor : TensorOf<[TF_Int16]>;
def TF_Int32Tensor : TensorOf<[TF_Int32]>;
def TF_Int64Tensor : TensorOf<[TF_Int64]>;
def TF_I32OrI64Tensor : TensorOf<[TF_I32OrI64]>;

  CHECK(SerializeToTString(s, output_tstring));
  TF_DeleteTensor(summary_tensor);
}

bool IsSameSize(TF_Tensor* tensor1, TF_Tensor* tensor2) {
  if (TF_NumDims(tensor1) != TF_NumDims(tensor2)) {
    return false;
  }
  for (int d = 0; d < TF_NumDims(tensor1); d++) {
    if (TF_Dim(tensor1, d) != TF_Dim(tensor2, d)) {
      return false;
    }
  }
  return true;
}

std::string SingleTag(TF_Tensor* tags) {

    auto* ret = emptyTensor->tensor;
    delete emptyTensor;
    return ret;
  }

  Tensor tensor;
  if (!tensor.CopyFrom(src, src.shape())) {
    return nullptr;
  }
  return new tensorflow::TensorInterface(std::move(tensor));
}

// Non-static for testing.
TF_Tensor* TF_TensorFromTensor(const tensorflow::Tensor& src, Status* status) {

  }];

  let arguments = (ins
    Arg<TF_Int32Tensor, [{int32 `Tensor`.}]>:$arr,
    Arg<TF_Int32Tensor, [{non-negative int32 scalar `Tensor`.}]>:$size,
    Arg<TensorOf<[TF_Float32, TF_Float64, TF_Int32, TF_Int64]>, [{is an int32, int64, float32, or float64 `Tensor` with the same
shape as `arr`, or a length-0 `Tensor`, in which case it acts as all weights
equal to 1.}]>:$weights
  );

  for (int component_index = 0; component_index < tensors_.size();
       ++component_index) {
    // TODO(allenl): Add a C API for summarizing tensors. Currently custom
    // devices limiting themselves to a C API (for ABI compatibility) would need
    // to implement summarization for component tensors themselves.
    ImmediateExecutionTensorHandle* component =
        tensorflow::unwrap(tensors_[component_index].get());

  DataType data_type() const { return data_type_; }
  Status status() const { return status_; }
  const Tensor& tensor() const { return tensor_; }

 private:
  Status status_;
  Output output_ = Output(Operation(nullptr), 0);
  Tensor tensor_;
  const std::string node_name_ = "";
  int32 index_ = 0;
  DataType data_type_ = DT_INVALID;
};