2.14.3-1

Add support for improved fault tolerance: non-blocking mode, new
init function with config, and ncclCommFinalize function.
Reintroduce collnet+chain algorithm, alongside collnet+direct.
Add LL protocol for intra-node P2P (on by default) and network
communication (off by default).
Use network instead of shared memory when performance is better.
Fix: wait for CUDA graph destroy before destroying comm with linked
graph resources.
Remove aggressive polling during enqueue.
Fix DMABUF fallback on MOFED 5.4 and earlier.

src/include/proxy.h

@@ -164,6 +164,7 @@ struct ncclProxyState {
   struct ncclSocket* listenSock;
   int stop;
   CUcontext cudaCtx;
+  int safeAbortFlag;
 
   // Used by main thread
   union ncclSocketAddress* peerAddresses;
@@ -183,6 +184,7 @@ struct ncclProxyConnection {
   struct ncclProxyArgs *proxyAppend;
   struct ncclProxyArgs **proxyAppendPtr;
   void* transportResources;
+  bool initFlag;
 };
 
 typedef ncclResult_t (*threadFunc_t)(struct ncclProxyArgs*);