MSCCL++ fix split path null deref (#1959)

* Add speriaswamy-amd to CODEOWNERS
* MSCCL++: fix split path null deref; key maps by parent ncclUniqueId
* removed no-op

[ROCm/rccl commit: 5bd5079de1]
Этот коммит содержится в:
Surya Periaswamy
2025-10-09 14:08:38 -05:00
коммит произвёл GitHub
родитель 8e5016ebfd
Коммит 014fae1b51
+4 -4
Просмотреть файл
@@ -2043,10 +2043,10 @@ static ncclResult_t ncclCommInitRankFunc(struct ncclAsyncJob* job_) {
comm->mscclppCompatible = true;
comm->mscclpp_threshold = job->parent->mscclpp_threshold;
comm->mscclpp_comm = job->parent->mscclpp_comm;
auto& mscclppUniqueId = mscclpp_uniqueIdMap[*job->commId];
mscclpp_uniqueIdMap[*job->commId] = mscclppUniqueId;
mscclpp_uniqueIdReverseMap[mscclppUniqueId].insert(*job->commId);
ncclCommToUniqueIdMap[comm] = *job->commId;
const ncclUniqueId& parentUniqueId = ncclCommToUniqueIdMap[job->parent];
auto& mscclppUniqueId = mscclpp_uniqueIdMap[parentUniqueId];
mscclpp_uniqueIdReverseMap[mscclppUniqueId].insert(parentUniqueId);
ncclCommToUniqueIdMap[comm] = parentUniqueId;
}
}
else