2.7.3-1

Add support for A100 GPU and related platforms. Add support for CUDA 11. Add support for send/receive operations (beta). [ROCm/rccl commit: 5949d96f36]
2020-05-12 14:40:18 -07:00
parent c43022b9d8
commit 89072c82e5
54 changed files with 2051 additions and 1254 deletions
@@ -0,0 +1,77 @@
+/*************************************************************************
+ * Copyright (c) 2016-2020, NVIDIA CORPORATION. All rights reserved.
+ *
+ * See LICENSE.txt for license information
+ ************************************************************************/
+
+#ifndef NCCL_PROXY_H_
+#define NCCL_PROXY_H_
+
+#include <pthread.h>
+
+enum ncclProxyOpState { ncclProxyOpNone, ncclProxyOpReady, ncclProxyOpProgress };
+
+struct ncclProxyArgs;
+typedef ncclResult_t (*proxyProgressFunc_t)(struct ncclProxyArgs*);
+
+struct ncclProxyArgs {
+  proxyProgressFunc_t progress;
+  struct ncclChannel* channel;
+  struct ncclConnector* connector;
+  int sliceSteps;
+  int chunkSteps;
+  int nsteps;
+  uint64_t opCount;
+  int protocol;
+  ncclDataType_t dtype;
+  ncclRedOp_t redOp;
+  int state;   // add component before this line -- it is left out during initialization
+
+  // Internal state
+  uint64_t head;
+  uint64_t tail;
+  uint64_t end;
+  void* requests[NCCL_STEPS];
+  int idle;
+
+  // Element linking
+  pthread_mutex_t mutex;
+  struct ncclProxyArgs* next;
+  struct ncclProxyArgs* nextPeer;
+};
+
+struct ncclProxyPool;
+struct ncclProxyState {
+  pthread_cond_t cond;
+  pthread_mutex_t mutex;
+  bool stop;
+  struct ncclProxyArgs* ops;
+  struct ncclProxyArgs* pool;
+  struct ncclProxyPool* pools;
+};
+
+typedef ncclResult_t (*threadFunc_t)(struct ncclProxyArgs*);
+
+enum proxyMode {
+  proxyRing = 0,
+  proxyFrom = 1,
+  proxyTo = 2
+};
+
+ncclResult_t ncclProxySaveColl(struct ncclProxyArgs* args, int pattern, int root, int nranks);
+ncclResult_t ncclProxySaveP2p(struct ncclInfo* info, struct ncclChannel* channel);
+ncclResult_t ncclProxyStart(struct ncclComm* comm);
+ncclResult_t ncclProxyCreate(struct ncclComm* comm);
+ncclResult_t ncclProxyDestroy(struct ncclComm* comm);
+
+#include <unistd.h>
+
+// Spin wait until func evaluates to true
+template<typename FUNC>
+inline void transportProxyWait(const FUNC& func) {
+  while (!func()) {
+    sched_yield();
+  }
+}
+
+#endif