kanshan/nccl-tests
1
0
Fork 0
mirror of https://github.com/NVIDIA/nccl-tests.git synced 2026-05-01 11:58:19 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Do not allocate memory for expected buffer if checking disabled

Browse Source 
This allows the tests to be run with larger buffers
This commit is contained in:
David Addison 2021-01-20 17:08:40 -08:00
parent 2f9bba9f20
commit 7677f3f608
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/common.cu
View File

@ -559,7 +559,7 @@ testResult_t threadLaunch(struct testThread* thread) {
testResult_t AllocateBuffs(void **sendbuff, size_t sendBytes, void **recvbuff, size_t recvBytes, void **expected, size_t nbytes, int nranks) {
CUDACHECK(cudaMalloc(sendbuff, nbytes));
CUDACHECK(cudaMalloc(recvbuff, nbytes));
CUDACHECK(cudaMalloc(expected, recvBytes));
if (datacheck) CUDACHECK(cudaMalloc(expected, recvBytes));
return testSuccess;
}
@ -764,7 +764,7 @@ testResult_t run() {
for (int i=0; i<nGpus*nThreads; i++) {
CUDACHECK(cudaSetDevice(localRank*nThreads*nGpus+i));
AllocateBuffs(sendbuffs+i, sendBytes, recvbuffs+i, recvBytes, expected+i, (size_t)maxBytes, nProcs*nThreads*nGpus);
TESTCHECK(AllocateBuffs(sendbuffs+i, sendBytes, recvbuffs+i, recvBytes, expected+i, (size_t)maxBytes, nProcs*nThreads*nGpus));
CUDACHECK(cudaStreamCreateWithFlags(streams+i, cudaStreamNonBlocking));
}
@ -868,7 +868,7 @@ testResult_t run() {
for (int i=0; i<nGpus*nThreads; i++) {
CUDACHECK(cudaFree(sendbuffs[i]));
CUDACHECK(cudaFree(recvbuffs[i]));
CUDACHECK(cudaFree(expected[i]));
if (datacheck) CUDACHECK(cudaFree(expected[i]));
}
CUDACHECK(cudaFreeHost(delta));