2.11.4-1

Add new API for creating a reduction operation which multiplies the input by a rank-specific scalar before doing an inter-rank summation (see: ncclRedOpCreatePreMulSum).
Improve CollNet (SHARP) performance of ncclAllReduce when captured in a CUDA Graph via user buffer registration.
Add environment variable NCCL_NET_PLUGIN="<suffix>" to allow user to choose among multiple NCCL net plugins by substituting into "libnccl-net-<suffix>.so".
Fix memory leak of NVB connections.
Fix topology detection of IB Virtual Functions (SR-IOV).


[ROCm/rccl commit: e11238b302]

projects/rccl/src/graph/xml.cc

@@ -625,7 +625,7 @@ ncclResult_t ncclTopoGetXmlFromGpu(struct ncclXmlNode* pciNode, nvmlDevice_t nvm
   NCCLCHECK(xmlGetAttrInt(gpuNode, "sm", &sm));
 
   struct ncclXmlNode* nvlNode = NULL;
-  NCCLCHECK(xmlGetSub(pciNode, "nvlink", &nvlNode));
+  NCCLCHECK(xmlGetSub(gpuNode, "nvlink", &nvlNode));
   if (nvlNode == NULL) {
     // NVML NVLink detection
     int maxNvLinks = (sm < 60) ? 0 : (sm < 70) ? 4 : (sm < 80) ? 6 : 12;