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

Merge pull request #67 from NVIDIA/big_buffers

Browse Source 
Do not allocate memory for expected buffer if checking disabled

[ROCm/rccl-tests commit: 0b30de583f]
This commit is contained in:
David Addison
2021-02-04 09:24:09 -08:00
committed by GitHub
parent 281348cba9 c62bde3272
commit a74716696b
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl-tests/src/common.cu
+3 -3
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));