Why is a mechanism like PXN in NCCL not used in low-latency kernel? It should be able to achieve even lower latency.

NCCL provides the PXN mechanism. During cross-node communication, if the GPUs are not of the same type, data can first be forwarded within the node to a GPU of the same type and then transferred via RDMA to the target node's GPU. On multi-rail networks, this can reduce the number of switch hops, theoretically achieving lower latency. In low-latency kernels, what is the reason for not adopting a similar mechanism?

Start by reading the low-latency kernel implementation and NCCL's PXN mechanism, focusing on cross-node communication, GPU type matching, forwarding, and RDMA paths. Compare the assumptions and document whether adopting a similar mechanism is feasible and what trade-offs or constraints determine the result.

Written by the indexing model from the issue text.