Multidimensional Slice in C++

markl · October 9, 2019, 5:08am

Is there a multidimensional slice in the C++ API?

I think there is a single dimensional one

pytorch/pytorch/blob/3ad1bbe16a3c1d6bb9566f09229afd63022a82df/aten/src/ATen/native/TensorShape.cpp#L655


    auto new_values = at::empty(vsize, values.options());
    for (int64_t k=0; k < nnz; k++) {
      new_values[k] = values[k].index_select(dim - sparse_dim, index);
    }
    return _sparse_coo_tensor_with_dims_and_tensors(
        sparse_dim, dense_dim, new_sizes, indices, new_values, self.options());


  }
}


Tensor slice(const Tensor& self, int64_t dim, int64_t start, int64_t end, int64_t step) {
  int64_t ndim = self.dim();
  if (ndim == 0) {
    AT_INDEX_ERROR("slice() cannot be applied to a 0-dim tensor.");
  }
  dim = maybe_wrap_dim(dim, ndim);
  auto sizes = self.sizes().vec();
  auto strides = self.strides().vec();
  // TODO: support negative strides
  TORCH_CHECK(step > 0, "slice step must be positive");
  if (start < 0) {

yf225 · October 9, 2019, 5:44am

We don’t have it now, but we will add it (aka. numpy-style indexing) by the end of this year.