if (TF_GetCode(s) != TF_OK) return;
  TFE_OpAddAttrs(op, attributes);
  TFE_OpSetDevice(op, dev->underlying_device.c_str(), s);
  if (TF_GetCode(s) != TF_OK) return;
  int num_inputs = TFE_OpGetFlatInputCount(original_op, s);
  if (TF_GetCode(s) != TF_OK) return;
  for (int j = 0; j < num_inputs; ++j) {
    TFE_TensorHandle* input = TFE_OpGetFlatInput(original_op, j, s);
    if (TF_GetCode(s) != TF_OK) return;

// already-finalized operation.
//
// Note that TFE_OpGetFlatInputCount and TFE_OpGetFlatInput operate on a flat
// sequence of inputs, unlike TFE_OpGetInputLength (for getting the length of a
// particular named input list, which may only be part of the op's inputs).
TF_CAPI_EXPORT extern int TFE_OpGetFlatInputCount(const TFE_Op* op,
                                                  TF_Status* status);

  NamedParallelDevice* named_device =
      reinterpret_cast<NamedParallelDevice*>(device_info);
  std::vector<MaybeParallelTensorUnowned> typed_inputs;
  int num_inputs = TFE_OpGetFlatInputCount(original_op, status);
  if (TF_GetCode(status) != TF_OK) return;
  typed_inputs.reserve(num_inputs);
  for (int i = 0; i < num_inputs; ++i) {
    TFE_TensorHandle* input = TFE_OpGetFlatInput(original_op, i, status);

      {reinterpret_cast<tensorflow::AbstractTensorHandle**>(
           tensorflow::unwrap(inputs)),
       static_cast<size_t>(num_inputs)});
}

extern int TFE_OpGetFlatInputCount(const TFE_Op* op, TF_Status* status) {
  return tensorflow::unwrap(op)->GetInputs().size();
}

extern TFE_TensorHandle* TFE_OpGetFlatInput(const TFE_Op* op, int index,

  TFE_OpSetDevice(ret, device, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddAttrs(ret, TFE_OpGetAttrs(other));
  int num_inputs = TFE_OpGetFlatInputCount(other, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  for (int input_index = 0; input_index < num_inputs; ++input_index) {