pytorch_geometric: Not found error for torch_sparse::ptr2ind in torchscript

❓ Questions & Help

I tried to use pytorch model with MessagePassing layer in C++ code. As described in pytorch_geometric documentation, I generate torch model with my own MP layer and successfully convert the model.

But in the process of executing C++ code, I face the error like below:

Unknown builtin op: torch_sparse::ptr2ind.
Could not find any similar ops to torch_sparse::ptr2ind. This op may not exist or may not be currently supported in TorchScript.
:
  File "/home/sr6/kyuhyun9.lee/env_ML/lib/python3.6/site-packages/torch_sparse/storage.py", line 166
        rowptr = self._rowptr
        if rowptr is not None:
            row = torch.ops.torch_sparse.ptr2ind(rowptr, self._col.numel())
                  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
            self._row = row
            return row
Serialized   File "code/__torch__/torch_sparse/storage.py", line 825
      if torch.__isnot__(rowptr, None):
        rowptr13 = unchecked_cast(Tensor, rowptr)
        row15 = ops.torch_sparse.ptr2ind(rowptr13, torch.numel(self._col))
                ~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
        self._row = row15
        _150, _151 = True, row15
'SparseStorage.row' is being compiled since it was called from 'SparseStorage.__init__'
  File "/home/sr6/kyuhyun9.lee/env_ML/lib/python3.6/site-packages/torch_sparse/storage.py", line 133
        if not is_sorted:
            idx = self._col.new_zeros(self._col.numel() + 1)
            idx[1:] = self._sparse_sizes[1] * self.row() + self._col
                                              ~~~~~~~~ <--- HERE
            if (idx[1:] < idx[:-1]).any():
                perm = idx[1:].argsort()
Serialized   File "code/__torch__/torch_sparse/storage.py", line 267
      idx = torch.new_zeros(self._col, [_29], dtype=None, layout=None, device=None, pin_memory=None)
      _30 = (self._sparse_sizes)[1]
      _31 = torch.add(torch.mul((self).row(), _30), self._col, alpha=1)
                                 ~~~~~~~~~~ <--- HERE
      _32 = torch.slice(idx, 0, 1, 9223372036854775807, 1)
      _33 = torch.copy_(_32, _31, False)
'SparseStorage.__init__' is being compiled since it was called from 'GINLayerJittable_d54f76.__check_input____1'
Serialized   File "code/__torch__/GINLayerJittable_d54f76.py", line 40
      pass
    return the_size
  def __check_input____1(self: __torch__.GINLayerJittable_d54f76.GINLayerJittable_d54f76,
      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~...  <--- HERE
    edge_index: __torch__.torch_sparse.tensor.SparseTensor,
    size: Optional[Tuple[int, int]]) -> List[Optional[int]]:

Aborted (core dumped)

Since I have no experience of pytorch jit, I cannot find any clue to solve this. How can I handle this error?