Merge remote-tracking branch 'nccl-tests/master' into HEAD

[ROCm/rccl-tests commit: 621dde544d]
This commit is contained in:
Wenkai Du
2024-03-01 17:39:14 +00:00
14 changed files with 184 additions and 335 deletions
+7 -12
View File
@@ -22,21 +22,16 @@ void HyperCubeGetCollByteCount(size_t *sendcount, size_t *recvcount, size_t *par
testResult_t HyperCubeInitData(struct threadArgs* args, ncclDataType_t type, ncclRedOp_t op, int root, int rep, int in_place) {
size_t sendcount = args->sendBytes / wordSize(type);
size_t recvcount = args->expectedBytes / wordSize(type);
int nranks = args->nProcs*args->nThreads*args->nGpus*args->nRanks;
int nranks = args->nProcs*args->nThreads*args->nGpus;
int k=0;
for (int i=0; i<args->nGpus; i++) {
HIPCHECK(hipSetDevice(args->gpus[i]));
for (int l=0; l<args->nRanks; l++) {
int rank = ((args->proc*args->nThreads + args->thread)*args->nGpus*args->nRanks + i*args->nRanks + l);
HIPCHECK(hipMemset(args->recvbuffs[k], 0, args->expectedBytes));
void* data = in_place ? ((char*)args->recvbuffs[k])+rank*args->sendBytes : args->sendbuffs[k];
TESTCHECK(InitData(data, sendcount, 0, type, ncclSum, 33*rep + rank, 1, 0));
for (int j=0; j<nranks; j++) {
TESTCHECK(InitData(((char*)args->expected[k])+args->sendBytes*j, sendcount, 0, type, ncclSum, 33*rep + j, 1, 0));
}
k++;
int rank = ((args->proc*args->nThreads + args->thread)*args->nGpus + i);
HIPCHECK(hipMemset(args->recvbuffs[i], 0, args->expectedBytes));
void* data = in_place ? ((char*)args->recvbuffs[i])+rank*args->sendBytes : args->sendbuffs[i];
TESTCHECK(InitData(data, sendcount, 0, type, ncclSum, 33*rep + rank, 1, 0));
for (int j=0; j<nranks; j++) {
TESTCHECK(InitData((char*)args->expected[i] + args->sendBytes*j, sendcount, 0, type, ncclSum, 33*rep + j, 1, 0));
}
HIPCHECK(hipDeviceSynchronize());
}