- The reduction primitive requires the source and destination tensors to have the same number of dimensions.
- Reduction dimensions are of size 1 in a destination tensor.
- The reduction primitive does not have a notion of forward or backward propagations.
- The memory format can be either specified explicitly or by dnnl::memory::format_tag::any (recommended), in which case the primitive will derive the most appropriate memory format based on the format of the source tensor.
- Whenever possible, avoid specifying different memory formats for source and destination tensors.