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

extending the unit-tests for multi-rank support

Browse Source
This commit is contained in:
Edgar
2022-03-15 18:53:52 -04:00
committed by Edgar Gabriel
parent 0336ffdf70
commit a87d61db2b
6 changed files with 90 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/common/TestBedChild.cpp
+18 -4
View File
@@ -126,6 +126,8 @@ namespace RcclUnitTesting
PIPE_READ(this->totalRanks);
PIPE_READ(this->rankOffset);
PIPE_READ(this->numCollectivesInGroup);
bool useMultiRankPerGpu;
PIPE_READ(useMultiRankPerGpu);
// Read the GPUs this child uses and prepare storage for collective args / datasets
int numGpus;
@@ -166,11 +168,23 @@ namespace RcclUnitTesting
break;
}
if (ncclCommInitRank(&this->comms[localRank], this->totalRanks, id, globalRank) != ncclSuccess)
if (useMultiRankPerGpu)
{
ERROR("Rank %d on child %d unable to call ncclCommInitRank\n", globalRank, this->childId);
status = TEST_FAIL;
break;
if (ncclCommInitRankMulti(&this->comms[localRank], this->totalRanks, id, globalRank, globalRank) != ncclSuccess)
{
ERROR("Rank %d on child %d unable to call ncclCommInitRankMulti\n", globalRank, this->childId);
status = TEST_FAIL;
break;
}
}
else
{
if (ncclCommInitRank(&this->comms[localRank], this->totalRanks, id, globalRank) != ncclSuccess)
{
ERROR("Rank %d on child %d unable to call ncclCommInitRank\n", globalRank, this->childId);
status = TEST_FAIL;
break;
}
}
}
if (status == TEST_SUCCESS)