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

2.9.9-1

Browse Source 
Fix crash when setting NCCL_MAX_P2P_NCHANNELS below nchannels.
Fix hang during sendrecv dynamic NVB connection establishment on
cubemesh topologies.
Add environment variable to only use SHARP on communicators beyond
a given number of ranks.
Add debug subsystem to trace memory allocations.
Fix compilation with TRACE=1. (Issue #505)


[ROCm/rccl commit: 3fec2fa5ee]
This commit is contained in:
Sylvain Jeaugey
2021-05-11 18:16:30 -07:00
parent 780273774a
commit 3dd3a1ca66
14 changed files with 142 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/graph.h
+1
View File
@@ -26,6 +26,7 @@ ncclResult_t ncclTopoComputePaths(struct ncclTopoSystem* system, struct ncclPeer
void ncclTopoFree(struct ncclTopoSystem* system);
ncclResult_t ncclTopoTrimSystem(struct ncclTopoSystem* system, struct ncclComm* comm);
ncclResult_t ncclTopoComputeP2pChannels(struct ncclComm* comm);
ncclResult_t ncclTopoGetNvbGpus(struct ncclTopoSystem* system, int rank, int* nranks, int** ranks);
// Query topology
ncclResult_t ncclTopoGetNetDev(struct ncclTopoSystem* system, int rank, struct ncclTopoGraph* graph, int channelId, int rr, int* net);