[Question] Multiple Vendor GPU Support

[jhlee508]

Does torchcomms support selecting multiple backends (e.g., NCCL, NCCLX, RCCL, RCCLX) at the same time, so that it can run P2P or All-Reduce communication between different vendor GPUs (e..g, NVIDIA GPU <-> AMD GPU) within a single operation?

If this is already supported, could you explain how it is possible to enable simultaneous use of multiple communication backends across different GPU vendors? If it is not currently supported, is there any plan or roadmap to add cross-vendor GPU communication capability (e.g., enabling collectives between NVIDIA and AMD GPUs) in the future?