2.9.9-1

Fix crash when setting NCCL_MAX_P2P_NCHANNELS below nchannels.
Fix hang during sendrecv dynamic NVB connection establishment on
cubemesh topologies.
Add environment variable to only use SHARP on communicators beyond
a given number of ranks.
Add debug subsystem to trace memory allocations.
Fix compilation with TRACE=1. (Issue #505)


[ROCm/rccl commit: 3fec2fa5ee]

projects/rccl/src/graph/xml.cc

@@ -469,7 +469,7 @@ ncclResult_t ncclTopoGetXmlFromSys(struct ncclXmlNode* pciNode, struct ncclXml*
     if (path == NULL) NCCLCHECK(getPciPath(busId, &path));
     NCCLCHECK(ncclTopoSetAttrFromSys(pciNode, path, "class", "class"));
   }
-  ncclDebugNoWarn = 1;
+  ncclDebugNoWarn = NCCL_GRAPH;
   NCCLCHECK(xmlGetAttrIndex(pciNode, "vendor", &index));
   if (index == -1) {
     if (path == NULL) getPciPath(busId, &path);