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

Adding UT_DEBUG_PAUSE to unit tests (#1653)

Browse Source
This commit is contained in:
gilbertlee-amd
2025-04-21 21:15:07 -06:00
committed by GitHub
parent ac8ec4c08c
commit ee85a70bb4
3 changed files with 24 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/common/TestBed.cpp
+19 -6
View File
@@ -120,6 +120,19 @@ namespace RcclUnitTesting
}
}
// If debugging is enabled, pause here to allow users to attach debugger
if (ev.debugPause) {
INFO("============================================================\n");
INFO(" Pausing for debug attach: (e.g. sudo rocgdb -p <PID>)\n");
INFO("============================================================\n");
for (int childId = 0; childId < this->numActiveChildren; ++childId) {
INFO(" Child %02d: processID: %d\n", childId, childList[childId]->pid);
}
INFO("============================================================\n");
INFO("<Press enter to continue>\n");
scanf("%*c");
}
// Determine number of unique GPUs being used.
std::set<int> unique_devices;
for (auto a: this->rankToDeviceMap)
@@ -252,7 +265,7 @@ namespace RcclUnitTesting
std::vector<int> rankList;
for (int i = 0; i < this->numActiveRanks; ++i)
if (rank == -1 || rank == i) rankList.push_back(i);
// Build list of groups this applies to (-1 for groupId means to set for all)
std::vector<int> groupList;
for (int i = 0; i < this->numGroupCalls; ++i)
@@ -287,7 +300,7 @@ namespace RcclUnitTesting
std::vector<int> rankList;
for (int i = 0; i < this->numActiveRanks; ++i)
if (rank == -1 || rank == i) rankList.push_back(i);
// Build list of groups this applies to (-1 for groupId means to set for all)
std::vector<int> groupList;
for (int i = 0; i < this->numGroupCalls; ++i)
@@ -311,7 +324,7 @@ namespace RcclUnitTesting
InteractiveWait("Finishing PrepareData");
}
void TestBed::ExecuteCollectives(std::vector<int> const &currentRanks, int const groupId,
void TestBed::ExecuteCollectives(std::vector<int> const &currentRanks, int const groupId,
bool const useHipGraph)
{
InteractiveWait("Starting ExecuteCollectives");
@@ -367,7 +380,7 @@ namespace RcclUnitTesting
std::vector<int> rankList;
for (int i = 0; i < this->numActiveRanks; ++i)
if (rank == -1 || rank == i) rankList.push_back(i);
// Build list of groups this applies to (-1 for groupId means to set for all)
std::vector<int> groupList;
for (int i = 0; i < this->numGroupCalls; ++i)
@@ -408,7 +421,7 @@ namespace RcclUnitTesting
if (groupId == -1 || groupId == i) groupList.push_back(i);
int const cmd = TestBedChild::CHILD_LAUNCH_GRAPHS;
for (auto currGroup : groupList)
for (auto currGroup : groupList)
{
for (int childId = 0; childId < this->numActiveChildren; ++childId)
{
@@ -550,7 +563,7 @@ namespace RcclUnitTesting
return ev.GetAllSupportedDataTypes();
}
std::vector<int> const TestBed::GetNumCollsPerGroup(int numCollectivesInGroup,
std::vector<int> const TestBed::GetNumCollsPerGroup(int numCollectivesInGroup,
int numGroupCalls)
{
return std::vector<int>(numGroupCalls, numCollectivesInGroup);