NCCL 2.26.3-1

Minimize the performance impact of the device kernel profiling support when
the profiler plugin is not loaded.

Reduce the overheads of CUDA graph capturing, which increased in NCCL
2.26.2 for large graphs.

Fix the exchange of enhanced connection establishment (ECE) options to
address potential slowdowns on networks utilizing RoCE.

Test if cuMem host allocations work and if not, disable them. Enabled by
default since NCCL 2.24 if the CUDA driver version is at least 12.6, such
allocations rely on NUMA support, which is by default not available under
Docker. We recommend invoking Docker with "--cap-add SYS_NICE" to enable
it.

Fix an initialization error when running with NCCL_NET_GDR_C2C=1 on
multiple MNNVL domains with non-uniform network configurations across
nodes.

Fix the printing of sub-seconds in the debug log when using a custom
NCCL_DEBUG_TIMESTAMP_FORMAT setting.

src/proxy.cc

@@ -9,7 +9,6 @@
 #include "collectives.h"
 #include "socket.h"
 #include "shmutils.h"
-#include "profiler.h"
 #define ENABLE_TIMER 0
 #include "timer.h"
 #include "profiler.h"
@@ -533,15 +532,21 @@ static ncclResult_t ncclLocalOpAppend(struct ncclComm* comm, struct ncclProxyCon
   return ncclSuccess;
 }
 
+static void incWorkCounter(struct ncclComm* comm, struct ncclProxyOp* op) {
+  op->workCounter = (op->incWorkCounter) ? ++comm->profiler.workCounter[op->channelId] : comm->profiler.workCounter[op->channelId];
+}
+
 static ncclResult_t SaveProxyProfiler(struct ncclComm* comm, struct ncclProxyOp* op, bool* justInquire) {
   struct ncclProxyConnector* proxyConn = (op->coll == ncclFuncRecv) ? &comm->profiler.recvProxyConn[op->channelId] : &comm->profiler.sendProxyConn[op->channelId];
-  if (justInquire) *justInquire = true;
-  else {
+  if (justInquire) {
+    *justInquire = true;
+    if (!comm->planner.persistent) incWorkCounter(comm, op);
+  } else {
     op->sendbuff = (uint8_t *)comm->profiler.workStarted;
     op->recvbuff = (uint8_t *)comm->profiler.workCompleted;
-    NCCLCHECK(ncclLocalOpAppend(comm, proxyConn, op));
     // Ensure that in graph capturing the proxy workCounter is incremented to keep up with kernel workCounter
-    op->workCounter += comm->profiler.workCounter[op->channelId];
+    if (comm->planner.persistent) incWorkCounter(comm, op);
+    NCCLCHECK(ncclLocalOpAppend(comm, proxyConn, op));
   }
   return ncclSuccess;
 }
@@ -696,9 +701,8 @@ ncclResult_t ncclProxySaveOp(struct ncclComm* comm, struct ncclProxyOp* op, bool
       NCCLCHECK(SaveProxy(comm, channel, op->pattern == ncclPatternSend ? proxySend : proxyRecv, op->root, op, 1, justInquire));
     } break;
   case ncclPatternProfiler: {
-      if (ncclProfilerNeedsProxy(comm, op)) {
-        NCCLCHECK(SaveProxyProfiler(comm, op, justInquire));
-      }
+      if (ncclProfilerNeedsProxy(comm, op)) NCCLCHECK(SaveProxyProfiler(comm, op, justInquire));
+      else incWorkCounter(comm, op);
     } break;
   }
   return ncclSuccess;