Merge remote-tracking branch 'rccl/develop' into 2.19.4

src/graph/paths.cc

@@ -863,7 +863,8 @@ ncclResult_t ncclTopoComputeP2pChannels(struct ncclComm* comm) {
   } else {
     // Round to next pow2 nChannelsPerPeer and nChannels
     comm->p2pnChannelsPerPeer = (ncclParamNChannelsPerPeer() == -2 ? nextPow2(minChannels) : ncclParamNChannelsPerPeer());
-    comm->p2pnChannelsPerPeer *= 2;
+    // Doubling P2P channels per peer on single node
+    if (comm->topo->nodes[GPU].count == comm->topo->nRanks && IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx94")) comm->p2pnChannelsPerPeer *= 2;
     comm->p2pnChannels = nextPow2(comm->p2pnChannels);
   }
 
@@ -874,9 +875,7 @@ ncclResult_t ncclTopoComputeP2pChannels(struct ncclComm* comm) {
   // fill the whole space of nChannels. To do so we mirror the bits in the
   // nChannels space.
   for (int c=0; c<comm->p2pnChannels; c++) {
-    int mirror = 0;
-    for (int b=1, mb=(comm->p2pnChannels>>1); b<comm->p2pnChannels; b<<=1, mb>>=1) if (c & b) mirror |= mb;
-    comm->p2pChannels[c] = mirror;
+    comm->p2pChannels[c] = mirrorBits(c, comm->p2pnChannels);
   }
   return ncclSuccess;
 }