// Setting a larger thread pool does not help with the Swift caller, as we use
  // a different TFE context for each thread of execution (for running graph
  // functions, and their send/recvs corountines).
  config.set_inter_op_parallelism_threads(1);

  TF_Buffer* ret = TF_NewBuffer();
  TF_CHECK_OK(MessageToBuffer(config, ret));
  return ret;
}

TF_Buffer* TF_CreateRunOptions(unsigned char enable_full_trace) {

#include "tensorflow/core/lib/llvm_rtti/llvm_rtti.h"
#include "tensorflow/core/platform/errors.h"

namespace tensorflow {
namespace gradients {
namespace {

// TODO(b/172558015): Using the pointer address as the identifier for the tensor
// may lead to collisions. Introduce another way to get a unique id for this
// tensor.
int64_t ToId(const AbstractTensorHandle* t) {

class GraphOperation;
class GraphTensor;

auto& kUnknownDim = shape_inference::InferenceContext::kUnknownDim;
auto& kUnknownRank = shape_inference::InferenceContext::kUnknownRank;

// GraphTensor wraps a `TF_Output`, i.e. a pointer to TF_Operation and the index
// into the list of outputs for the operation.
class GraphTensor : public TracingTensorHandle {
 public:
  explicit GraphTensor(TF_Output output, TF_Graph* graph)

}

// This function will block till the operation that produces `h` has
// completed. This is only valid on local TFE_TensorHandles. Returns the size in
// bytes of the memory pointed to by the device pointer returned above.
size_t TFE_TensorHandleDeviceMemorySize(TFE_TensorHandle* h,
                                        TF_Status* status) {
  if (h == nullptr) {