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

do not use internal stream (#79)

Browse Source 
[ROCm/rccl commit: 754ed213cc]
This commit is contained in:
Jeff Daily
2019-06-12 15:26:59 -07:00
committed by gilbertlee-amd
parent 87d5441552
commit 53b1ca1d7f
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/init.cu
+2 -2
View File
@@ -45,7 +45,7 @@ FILE *ncclDebugFile = stdout;
std::chrono::high_resolution_clock::time_point ncclEpoch;
#endif
#if CUDART_VERSION >= 9200
#if CUDART_VERSION >= 9200 || defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#define NCCL_GROUP_CUDA_STREAM 0 // CGMD: CUDA 9.2,10.X Don't need to use an internal CUDA stream
#else
#define NCCL_GROUP_CUDA_STREAM 1 // CGMD: CUDA 9.0,9.1 Need to use an internal CUDA stream
@@ -250,7 +250,7 @@ static ncclResult_t commAlloc(ncclComm_t* comret, int ndev, int rank) {
comm->doneEvent = doneEvent;
comm->llThreshold = ncclParamLlThreshold();
comm->checkPointers = ncclParamCheckPointers() == 1 ? true : false;
#if CUDART_VERSION >= 9200
#if CUDART_VERSION >= 9200 || defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
comm->groupCudaStream = ncclParamGroupCudaStream();
#else
// Don't allow the user to overload the default setting in older CUDA builds