2.9.8-1

Fix memory leaks.
Fix crash in bootstrap error case.
Fix Collnet clean-up issue.
Make PCI switch vendor/device optional for XML injection.
Add support for nvidia-peermem module.


[ROCm/rccl commit: ca8485b0d0]

projects/rccl/src/graph/connect.cc

@@ -212,6 +212,7 @@ static ncclResult_t connectCollNet(struct ncclComm* comm, struct ncclTopoGraph*
     sprintf(line+strlen(line), "headRank %d out %d shift %d", channel->collTree.headRank, channel->collTree.out, channel->collTree.shift);
     INFO(NCCL_GRAPH, "%s", line);
   }
+  free(heads);
   return ncclSuccess;
 }
 

projects/rccl/src/graph/search.cc

@@ -764,7 +764,7 @@ search:
     for (int g=0; g<ngpus; g++) {
       printf("%d ", graph->intra[c*ngpus+g]);
     }
-    printf("[%d %d]", graph->inter[0], graph->inter[1]);
+    printf("[%d %d]", graph->inter[c*2+0], graph->inter[c*2+1]);
     printf("\n");
   }
 #endif

projects/rccl/src/graph/xml.cc

@@ -469,26 +469,28 @@ ncclResult_t ncclTopoGetXmlFromSys(struct ncclXmlNode* pciNode, struct ncclXml*
     if (path == NULL) NCCLCHECK(getPciPath(busId, &path));
     NCCLCHECK(ncclTopoSetAttrFromSys(pciNode, path, "class", "class"));
   }
+  ncclDebugNoWarn = 1;
   NCCLCHECK(xmlGetAttrIndex(pciNode, "vendor", &index));
   if (index == -1) {
-    if (path == NULL) NCCLCHECK(getPciPath(busId, &path));
-    NCCLCHECK(ncclTopoSetAttrFromSys(pciNode, path, "vendor", "vendor"));
+    if (path == NULL) getPciPath(busId, &path);
+    if (path) ncclTopoSetAttrFromSys(pciNode, path, "vendor", "vendor");
   }
   NCCLCHECK(xmlGetAttrIndex(pciNode, "device", &index));
   if (index == -1) {
-    if (path == NULL) NCCLCHECK(getPciPath(busId, &path));
-    NCCLCHECK(ncclTopoSetAttrFromSys(pciNode, path, "device", "device"));
+    if (path == NULL) getPciPath(busId, &path);
+    if (path) ncclTopoSetAttrFromSys(pciNode, path, "device", "device");
   }
   NCCLCHECK(xmlGetAttrIndex(pciNode, "subsystem_vendor", &index));
   if (index == -1) {
-    if (path == NULL) NCCLCHECK(getPciPath(busId, &path));
-    NCCLCHECK(ncclTopoSetAttrFromSys(pciNode, path, "subsystem_vendor", "subsystem_vendor"));
+    if (path == NULL) getPciPath(busId, &path);
+    if (path) ncclTopoSetAttrFromSys(pciNode, path, "subsystem_vendor", "subsystem_vendor");
   }
   NCCLCHECK(xmlGetAttrIndex(pciNode, "subsystem_device", &index));
   if (index == -1) {
-    if (path == NULL) NCCLCHECK(getPciPath(busId, &path));
-    NCCLCHECK(ncclTopoSetAttrFromSys(pciNode, path, "subsystem_device", "subsystem_device"));
+    if (path == NULL) getPciPath(busId, &path);
+    if (path) ncclTopoSetAttrFromSys(pciNode, path, "subsystem_device", "subsystem_device");
   }
+  ncclDebugNoWarn = 0;
   NCCLCHECK(xmlGetAttrIndex(pciNode, "link_speed", &index));
   if (index == -1) {
     if (path == NULL) NCCLCHECK(getPciPath(busId, &path));