AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

2.20.3-1

Browse Source 
Add support for alternating rings, allow for cross-nic rings without
cross-rail communication.
Add support for user buffer registration for network send/recv.
Optimize aggregated operations to better utilize all channels.
Add flattening for BCM PCI gen5 switches.
Add support for inter-node NVLink communication
Add support for port fusion in NET/IB.
Add support for ReduceScatter and AllGather using Collnet.
Update net API to v8.
Fix hang during A2A connection.


[ROCm/rccl commit: b6475625fb]
This commit is contained in:
Sylvain Jeaugey
2024-02-05 05:06:02 -08:00
parent 382ae08419
commit 2ab8a3a750
74 changed files with 4632 additions and 2165 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/graph.h
+4 -1
View File
@@ -33,6 +33,7 @@ int ncclTopoPathAllNVLink(struct ncclTopoSystem* system);
// Query topology
ncclResult_t ncclTopoGetNetDev(struct ncclComm* comm, int rank, struct ncclTopoGraph* graph, int channelId, int peerRank, int* net, int* proxyRank);
ncclResult_t ncclTopoCheckP2p(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* p2p, int *read, int* intermediateRank);
ncclResult_t ncclTopoCheckMNNVL(struct ncclTopoSystem* system, struct ncclPeerInfo* info1, struct ncclPeerInfo* info2, int* ret);
ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* topo, int64_t busId, int netDev, int read, int* useGdr);
ncclResult_t ncclTopoNeedFlush(struct ncclTopoSystem* system, int64_t busId, int* flush);
ncclResult_t ncclTopoCheckNet(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* net);
@@ -53,10 +54,11 @@ ncclResult_t ncclTopoGetCpuAffinity(struct ncclTopoSystem* system, int rank, cpu
#define NCCL_TOPO_CPU_TYPE_YONGFENG 1
ncclResult_t ncclTopoCpuType(struct ncclTopoSystem* system, int* arch, int* vendor, int* model);
ncclResult_t ncclTopoGetGpuCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetNvsCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetNetCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetNvsCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetLocalNet(struct ncclTopoSystem* system, int rank, int channelId, int* id);
ncclResult_t ncclTopoGetLocalGpu(struct ncclTopoSystem* system, int net, int* gpuIndex);
ncclResult_t getLocalNetCountByBw(struct ncclTopoSystem* system, int gpu, int *count);
#define NCCL_TOPO_MAX_NODES 256
@@ -102,6 +104,7 @@ struct ncclTopoRanks {
int treeToChild0[MAXCHANNELS];
int treeToChild1[MAXCHANNELS];
int nvlsHeads[MAXCHANNELS];
int nvlsHeadNum;
};
ncclResult_t ncclTopoPreset(struct ncclComm* comm, struct ncclTopoGraph** graphs, struct ncclTopoRanks* topoRanks);