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

Wrap ncclCommWindowRegister() calls within ncclGroup

Browse Source 
[ROCm/rccl-tests commit: e7c8825b0b]
This commit is contained in:
David Addison
2025-06-03 10:36:53 -07:00
parent 3b79e1a05c
commit 0ae7c8cbf4
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl-tests/src/common.cu
+4
View File
@@ -659,6 +659,7 @@ testResult_t threadInit(struct threadArgs* args) {
}
NCCLCHECK(ncclGroupEnd());
#if NCCL_VERSION_CODE >= NCCL_VERSION(2,19,0)
NCCLCHECK(ncclGroupStart());
void **sendRegHandles = (local_register) ? (void **)malloc(sizeof(*sendRegHandles)*args->nGpus) : NULL;
void **recvRegHandles = (local_register) ? (void **)malloc(sizeof(*recvRegHandles)*args->nGpus) : NULL;
for (int i=0; i<args->nGpus; i++) {
@@ -673,6 +674,7 @@ testResult_t threadInit(struct threadArgs* args) {
if (local_register) NCCLCHECK(ncclCommRegister(args->comms[i], args->recvbuffs[i], args->maxbytes, &recvRegHandles[i]));
}
}
NCCLCHECK(ncclGroupEnd());
#endif
TESTCHECK(threadRunTests(args));
@@ -1124,6 +1126,7 @@ testResult_t run() {
NCCLCHECK(ncclGroupEnd());
}
#if NCCL_VERSION_CODE >= NCCL_VERSION(2,19,0)
NCCLCHECK(ncclGroupStart());
sendRegHandles = (local_register) ? (void **)malloc(sizeof(*sendRegHandles)*nThreads*nGpus) : NULL;
recvRegHandles = (local_register) ? (void **)malloc(sizeof(*recvRegHandles)*nThreads*nGpus) : NULL;
for (int i=0; i<nGpus*nThreads; i++) {
@@ -1138,6 +1141,7 @@ testResult_t run() {
if (local_register) NCCLCHECK(ncclCommRegister(comms[i], recvbuffs[i], maxBytes, &recvRegHandles[i]));
}
}
NCCLCHECK(ncclGroupEnd());
#endif
}