gfx12 Disable ll protocol (#1268)

src/graph/tuning.cc

@@ -54,7 +54,7 @@ ncclResult_t parseList(const char* str, const char* elems[], int nelems, int* li
 
 // Latencies in us, Bandwidths in GB/s
 // Tree { LL, LL128, Simple } , Ring { LL, LL128, Simple }
-static const float baseLat  [NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS] = { 
+static const float baseLat  [NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS] = {
        { 12.0, 12.0, 17.0 }, { 12.0, 12.0, 17.0 },   // Tree, Ring
        { 12.0, 12.0, 17.0 }, { 12.0, 12.0, 17.0 },   // Collnet Direct, Chain
        {    0,    0,    0 }, {    0,    0,    0 }};  // NVLS, NVLS Tree
@@ -372,7 +372,7 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
 
         if (a == NCCL_ALGO_NVLS) bw = std::min(graphs[a]->bwIntra, graphs[a]->bwInter);
         if (a == NCCL_ALGO_NVLS_TREE) bw = std::min(graphs[a]->bwIntra, nNodes <= 2 ? graphs[a]->bwInter : graphs[a]->bwInter/2);
-        
+
         // Various model refinements
 #if defined(__HIP_PLATFORM_AMD__) || defined(__HCC__) || defined(__HIPCC__)
         if (nNodes <= 2)
@@ -507,8 +507,8 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
   }
 
   for (int c=0; c<NCCL_NUM_FUNCTIONS; c++) for (int a=0; a<NCCL_NUM_ALGORITHMS; a++) for (int p=0; p<NCCL_NUM_PROTOCOLS; p++) {
-    // Disable LL protocol on gfx11xx
-    int pEnable = protoEnable[p];
+    // Disable LL protocol on gfx12xx
+    int pEnable = (p == NCCL_PROTO_LL && IsArchMatch(comm->topo->nodes[GPU].nodes[0].gpu.gcn, "gfx12")) ? 0 : protoEnable[p];
     if (pEnable == 2 && p == NCCL_PROTO_LL128) {
 #if defined(__HIP_PLATFORM_AMD__) || defined(__HCC__) || defined(__HIPCC__)
 #if defined(ENABLE_LL128)