Possible features for hardware offload

Apart from Checksum offload, I am trying to identify what other parts
of the network stack can be potentially offloaded to hardware with the
goal of reducing the overall CPU utilization for very high data rates.
I am looking for high level data points which I can further research
into.

Few Additional questions
Is there documentation available for GRO feature, that can help me
understand the architecture and also the kernel implementation?
Looking for pointers to start. Same question for TCP segmentation
offload feature.
Is there a kernel supported way of offloading netfilter filtering to hardware?

Regards,
Sharat