Setup collectives threshold for enabling intranet (#387)

* Setup collectives threshold for enabling intranet

* Use separate operation counters for coll and p2p

src/include/comm.h

@@ -98,6 +98,8 @@ struct ncclComm {
   uint64_t opCount;
   // Collective operation counter
   uint64_t collOpCount;
+  // P2P operation counter
+  uint64_t p2pOpCount;
 
   // Channels for collectives
   int nChannels;
@@ -127,6 +129,7 @@ struct ncclComm {
 
   // Flags for enable P2P NET
   uint32_t *p2pNet;
+  uint32_t useIntraNet;
 
   // Device side of the communicator
   struct ncclDevComm *devComm;

src/include/devcomm.h

@@ -200,6 +200,7 @@ struct ncclWorkElem {
       uint32_t root;
       uint8_t bid;
       uint8_t nChannels;
+      uint8_t connIndex;
     } coll;
     struct {
       size_t sendCount;

src/include/proxy.h

@@ -60,6 +60,7 @@ struct ncclProxyArgs {
 
   int idle;
   uint64_t hdp_flushed;
+  uint8_t connIndex;
 
   // Element linking
   pthread_mutex_t mutex;