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

Merge remote-tracking branch 'nccl/master' into develop

Browse Source
This commit is contained in:
Wenkai Du
2023-06-21 20:54:24 -07:00
parent 0c77c66221 ea38312273
commit abd0615351
83 changed files with 5798 additions and 4090 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/include/graph.h
+9 -8
View File
@@ -59,9 +59,11 @@ ncclResult_t ncclTopoGetCpuAffinity(struct ncclTopoSystem* system, int rank, cpu
#define NCCL_TOPO_CPU_TYPE_ROME 4
#define NCCL_TOPO_CPU_TYPE_YONGFENG 1
ncclResult_t ncclTopoCpuType(struct ncclTopoSystem* system, int* arch, int* vendor, int* model);
ncclResult_t ncclTopoGetNetCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetGpuCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetNvsCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetLocalNet(struct ncclTopoSystem* system, int rank, int* id);
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);
#define NCCL_TOPO_MAX_NODES 256
@@ -72,6 +74,7 @@ ncclResult_t ncclTopoSearchInit(struct ncclTopoSystem* system);
#define NCCL_TOPO_PATTERN_SPLIT_TREE 2 // Spread NIC traffic between two GPUs (Tree parent on first GPU, tree children on the second GPU)
#define NCCL_TOPO_PATTERN_TREE 3 // All NIC traffic going to/from the same GPU
#define NCCL_TOPO_PATTERN_RING 4 // Ring
#define NCCL_TOPO_PATTERN_NVLS 5 // NVLS+SHARP and NVLS+Tree
struct ncclTopoGraph {
// Input / output
int id; // ring : 0, tree : 1, collnet : 2
@@ -108,18 +111,16 @@ struct ncclTopoRanks {
int treeToParent[MAXCHANNELS];
int treeToChild0[MAXCHANNELS];
int treeToChild1[MAXCHANNELS];
int nvlsHeads[MAXCHANNELS];
};
ncclResult_t ncclTopoPreset(struct ncclComm* comm,
struct ncclTopoGraph* treeGraph, struct ncclTopoGraph* ringGraph, struct ncclTopoGraph* collNetGraph,
struct ncclTopoRanks* topoRanks);
ncclResult_t ncclTopoPreset(struct ncclComm* comm, struct ncclTopoGraph** graphs, struct ncclTopoRanks* topoRanks);
ncclResult_t ncclTopoPostset(struct ncclComm* comm, int* firstRanks, int* treePatterns,
struct ncclTopoRanks** allTopoRanks, int* rings, struct ncclTopoGraph* collNetGraph, int nc);
struct ncclTopoRanks** allTopoRanks, int* rings, struct ncclTopoGraph** graphs, int nc);
ncclResult_t ncclTreeBasePostset(struct ncclComm* comm, struct ncclTopoGraph* treeGraph);
ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCompCap, struct ncclTopoGraph* treeGraph, struct ncclTopoGraph* ringGraph, struct ncclTopoGraph* collNetGraph);
ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCompCap, struct ncclTopoGraph** graphs);
#include "info.h"
ncclResult_t ncclTopoGetAlgoTime(struct ncclInfo* info, int algorithm, int protocol, int numPipeOps, float* time);