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 
[ROCm/rccl commit: 4f0e223db4]
This commit is contained in:
Wenkai Du
2022-10-20 15:40:03 +00:00
parent cf1d4a62e8 8761a6c2fc
commit 36e5e02e46
51 changed files with 2139 additions and 1081 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/graph.h
+4 -2
View File
@@ -37,6 +37,7 @@ ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* topo, int64_t busId, int ne
#define MAX_XGMI_INTER_GPUS 4
ncclResult_t ncclTopoGetIntraNetDev(struct ncclTopoSystem* system, int rank, struct ncclTopoGraph* graph, int channelId, int type, int* dev);
ncclResult_t ncclTopoGetLinkType(struct ncclTopoSystem* system, int cudaDev1, int cudaDev2, bool* isXGMI, int maxInter=MAX_XGMI_INTER_GPUS, int nInter=0, int *inter=nullptr);
ncclResult_t ncclTopoCheckNet(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* net);
int ncclPxnDisable(struct ncclComm* comm);
ncclResult_t ncclTopoGetPxnRanks(struct ncclComm* comm, int** intermediateRanks, int* nranks);
ncclResult_t ncclTopoGetLocalRank(struct ncclTopoSystem* system, int rank, int* localRank);
@@ -57,6 +58,7 @@ 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 ncclTopoGetNetCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetNvsCount(struct ncclTopoSystem* system, int* count);
ncclResult_t ncclTopoGetLocalNet(struct ncclTopoSystem* system, int rank, int* id);
#define NCCL_TOPO_MAX_NODES 256
@@ -78,8 +80,8 @@ struct ncclTopoGraph {
int maxChannels;
// Output
int nChannels;
float speedIntra;
float speedInter;
float bwIntra;
float bwInter;
float latencyInter;
int typeIntra;
int typeInter;