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

Add GPU memory usage tracker (#326)

Browse Source 
[ROCm/rccl commit: f60b76c67a]
This commit is contained in:
Wenkai Du
2021-03-06 20:32:30 -08:00
committed by GitHub
parent b7253710ca
commit 7374c512d9
3 changed files with 25 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/group.cc
+2
View File
@@ -15,6 +15,7 @@ thread_local pthread_t ncclGroupThreads[MAX_ASYNC_OPS];
thread_local int ncclGroupIndex = 0;
thread_local int ncclGroupMode = 0;
thread_local ncclResult_t ncclGroupError = ncclSuccess;
extern struct allocationTracker allocTracker[];
bool ncclAsyncMode() {
return ncclGroupMode > 0;
@@ -205,6 +206,7 @@ ncclResult_t ncclGroupEnd() {
WARN("Error waiting for pthread_join : %s", strerror(errno));
return ncclSystemError;
}
INFO(NCCL_INIT, "comm %p rank %d total %ld bytes - P2P preconnect COMPLETE", args->coll.comm, args->coll.comm->rank, allocTracker[args->coll.comm->cudaDev].totalAllocSize);
NCCLCHECKGOTO(args->ret, ret, end);
args->coll.comm->connect = 0;
}