Adding performance collection feature in rccl_replayer, and updating MSCCL logging and replayer parsing (#1265)
* Adding performance collection feature in rccl_replayer, and updating MSCCL logging and replayer parsing * Performance collection feature in rccl_replayer, and updating MSCCL logging and replayer parsing
Этот коммит содержится в:
@@ -377,9 +377,9 @@ static ncclResult_t mscclSaveCountsAndDispls(struct mscclSavedSchedulerParam* pa
|
||||
static ncclResult_t mscclRunSavedParams() {
|
||||
mscclThreadLocalStatus& threadLocalStatus = mscclGetThreadLocalStatus();
|
||||
for (auto& param : threadLocalStatus.savedSchedulerParams) {
|
||||
INFO(NCCL_COLL,"%s: opCount %lx sendbuff %p recvbuff %p count %zi datatype %d op %d root %d comm %p [nranks=%d] stream %p",
|
||||
INFO(NCCL_COLL,"%s: opCount %lx sendbuff %p recvbuff %p count %zi datatype %d op %d root %d comm %p [nranks=%d] stream %p task %d globalrank %d",
|
||||
mscclFuncNames[param.p.func], param.p.opCount, param.p.sendBuff, param.p.recvBuff, param.p.count,
|
||||
param.p.dataType, param.p.op, param.p.root, param.comm, param.p.nRanks, param.stream);
|
||||
param.p.dataType, param.p.op, param.p.root, param.comm, param.p.nRanks, param.stream, param.comm->tasks.nTasksP2p + param.comm->tasks.nTasksColl, param.comm->localRankToRank[param.comm->localRank]);
|
||||
|
||||
NCCLCHECK(mscclRunAlgo(
|
||||
param.p.sendBuff, param.p.sendCounts, param.p.sDisPls,
|
||||
|
||||
Ссылка в новой задаче
Block a user