2.3.5-5

Add support for inter-node communication using sockets and InfiniBand/RoCE.
Improve latency.
Add support for aggregation.
Improve LL/regular tuning.
Remove tests as those are now at github.com/nvidia/nccl-tests .


[ROCm/rccl commit: f93fe9bfd9]

projects/rccl/src/include/bootstrap.h

@@ -0,0 +1,18 @@
+/*************************************************************************
+ * Copyright (c) 2015-2018, NVIDIA CORPORATION. All rights reserved.
+ *
+ * See LICENSE.txt for license information
+ ************************************************************************/
+
+#ifndef NCCL_BOOTSTRAP_H_
+#define NCCL_BOOTSTRAP_H_
+
+#include "nccl.h"
+
+ncclResult_t bootstrapCreateRoot(ncclUniqueId* commId, bool idFromEnv);
+ncclResult_t bootstrapGetUniqueId(ncclUniqueId* out);
+ncclResult_t bootstrapInit(ncclUniqueId* id, int rank, int nranks, void** commState);
+ncclResult_t bootstrapAllGather(void* commState, void* allData, int size);
+ncclResult_t bootstrapRingExchange(void* commState, void* prevNextData, int prev, int next, int size);
+ncclResult_t bootstrapClose(void* commState);
+#endif