Could someone please explain me the following

The backward pass is implemented only for src.shape == index.shape.

Does this condition for torch.Tensor.scatter requires that src and index must have same number of dimension or their size must be be same? According to the condition, their shape,i.e., size must be same but this is not meaningful. The src stores features with N-D, but index stores 1-D/2-D integer indices.

Thank you