Better error message for old NCCL.

Author: abergeron

src/loaders/libnccl.c

@@ -40,6 +40,9 @@ int load_libnccl(error *e) {
 
   #include "libnccl.fn"
 
+  if (ga_func_ptr(lib, "ncclGroupStart", e) == NULL)
+    return error_set(e, GA_LOAD_ERROR, "Found NCCL 1.0 but NCCL 2.0 required");
+
   loaded = 1;
   return GA_NO_ERROR;
 }

src/loaders/libnccl.fn

@@ -9,5 +9,3 @@ DEF_PROC(ncclResult_t, ncclAllReduce, (const void* sendbuff, void* recvbuff, siz
 DEF_PROC(ncclResult_t, ncclReduceScatter, (const void* sendbuff, void* recvbuff, size_t recvcount, ncclDataType_t datatype, ncclRedOp_t op, ncclComm_t comm, cudaStream_t stream));
 DEF_PROC(ncclResult_t, ncclBcast, (void* buff, size_t count, ncclDataType_t datatype, int root, ncclComm_t comm, cudaStream_t stream ));
 DEF_PROC(ncclResult_t, ncclAllGather, (const void* sendbuff, void* recvbuff, size_t sendcount, ncclDataType_t datatype, ncclComm_t comm, cudaStream_t stream));
-// We don't need this but we use it as a sentinel to prevent nccl 1.0 from loading.
-DEF_PROC(ncclResult_t, ncclGroupStart, ());