Disable the __threadfence on the sender side of the simple protocol when possible. (#1830)

Leverages the traits of extended-scope fine-grain memory to get rid of a device-scope acquire-release fence.  This improves throughput for single node workloads on gfx942 and gfx950 for some input sizes (e.g., ~32 MiB to about 256 MiB) when using the simple protocol.  Multinode workloads on MI300X see a smaller but statistically significant uplift for some message sizes.  Runtime disablement is supported via setting the environment variable RCCL_GFX942_CHEAP_FENCE_ON to 0.

[ROCm/rccl commit: 1aa2570b48]

projects/rccl/src/init.cc

@@ -1369,13 +1369,23 @@ static ncclResult_t initTransportsRank(struct ncclComm* comm, struct ncclComm* p
     allGather3Data[rank].nc = std::max(allGather3Data[rank].nc, 4/ringGraph->nChannels);
   if (ringGraph->nChannels > MAXCHANNELS/2)
     allGather3Data[rank].nc = 1;
+  comm -> gfx942CheapFenceOff = 1;
+  #ifdef HIP_UNCACHED_MEMORY
+  if(!rcclParamGfx942CheapFenceOff()){
+    if(IsArchMatch(comm->topo->nodes[GPU].nodes[idx].gpu.gcn, "gfx942")){
+      comm -> gfx942CheapFenceOff = 0;
+    }
+    else if(IsArchMatch(comm->topo->nodes[GPU].nodes[idx].gpu.gcn, "gfx950")){
+      comm -> gfx942CheapFenceOff = nNodes > 1;
+    }
+  }
+  #endif
   if (IsArchMatch(comm->topo->nodes[GPU].nodes[idx].gpu.gcn, "gfx942")) {
     // Multi-node MI300A
     int managed = 0;
     CUDACHECK(hipDeviceGetAttribute(&managed, hipDeviceAttributeDirectManagedMemAccessFromHost, 0));
     // RCCL: Only use one slice per primitive on some single node gfx9xx systems
     comm->rcclUseOneSlice = !managed && nNodes == 1;
-    comm->gfx942CheapFenceOff = rcclParamGfx942CheapFenceOff();
     if (managed && nNodes > 1) {
       // This forces the minimum channels to 24
       allGather3Data[rank].nc = 6;