Fix collective trace when rccl is configured (#1056)

* Fix collective trace when rccl is configured

src/include/collectives.h

@@ -21,6 +21,7 @@ struct ncclDevRedOpFull {
 
 #define FUNC_INDEX_P2P 1015
 #define FUNC_INDEX_ALLTOALL_PIVOT 675
+#define FUNC_INDEX_TOTAL 1026  // Total number of functions that goes into librccl.so index in host_table.cc
 
 #define NCCL_FUNC_NAME(func, algo, proto, devredop, type) \
   ncclFunction_##func##_##algo##_##proto##_##devredop##_##type

src/include/devcomm.h

@@ -21,6 +21,9 @@
 #define NCCL_NUM_FUNCTIONS 5 // SendRecv and AllToAllPivot not included for now
 typedef enum { ncclFuncBroadcast, ncclFuncReduce, ncclFuncAllGather, ncclFuncReduceScatter, ncclFuncAllReduce, ncclFuncSendRecv, ncclFuncSend, ncclFuncRecv, ncclFuncAllToAllPivot, ncclNumFuncs} ncclFunc_t;
 extern const char* ncclFuncStr[NCCL_NUM_FUNCTIONS+2];
+#ifdef ENABLE_COLLTRACE
+extern const char* funcNames[FUNC_INDEX_TOTAL];
+#endif
 
 #define NCCL_NUM_ALGORITHMS 6 // Tree/Ring/CollNet*
 #define NCCL_ALGO_TREE 0