GPUDirect on 2 IOH : Feasible?

I have an 8 gpu system running with a lot of inter-data exchange, on 2 IOHS. The following output displays which device can access the next device via P2P
af_multi7.cpp(672):0 ← 1 Can access? 1
af_multi7.cpp(672):1 ← 2 Can access? 1
af_multi7.cpp(672):2 ← 3 Can access? 1
af_multi7.cpp(672):3 ← 4 Can access? 0
af_multi7.cpp(672):4 ← 5 Can access? 1
af_multi7.cpp(672):5 ← 6 Can access? 1
af_multi7.cpp(672):6 ← 7 Can access? 1
af_multi7.cpp(672):7 ← 0 Can access? 0

Since it’s a 2 IOH system, the links 4 < 3 and 0 < 7 are unsupported at the moment, and the transfers will only run as quickly as the slowest link (which would be one of these 2). I was wondering firstly whether it is technically possible for P2P to in the future be implemented to support multiple IOHs, and whether this has actually been planned?

Thanks.