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

Replace CUDA_VERSION by CUDART_VERSION

Browse Source 
[ROCm/rccl commit: c244b51ae7]
This commit is contained in:
Sylvain Jeaugey
2018-12-13 15:22:17 -08:00
parent 6a93dca51b
commit 4e32c7a19a
4 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/init.cu
+3 -3
View File
@@ -42,7 +42,7 @@ FILE *ncclDebugFile = stdout;
std::chrono::high_resolution_clock::time_point ncclEpoch;
#endif
#if CUDA_VERSION >= 9200
#if CUDART_VERSION >= 9200
#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
@@ -229,7 +229,7 @@ static ncclResult_t commAlloc(ncclComm_t* comret, int ndev, int rank) {
comm->doneEvent = doneEvent;
comm->llThreshold = ncclParamLlThreshold();
comm->checkPointers = ncclParamCheckPointers() == 1 ? true : false;
#if CUDA_VERSION >= 9200
#if CUDART_VERSION >= 9200
comm->groupCudaStream = ncclParamGroupCudaStream();
#else
// Don't allow the user to overload the default setting in older CUDA builds
@@ -473,7 +473,7 @@ ncclResult_t ncclCommSetIntra(struct ncclComm* comm, int rank, int ranks, struct
}
if (comm->launchMode == ncclComm::GROUP) {
CUDACHECK(cudaStreamCreateWithFlags(&comm->groupStream, cudaStreamNonBlocking));
#if CUDA_VERSION >= 9000
#if CUDART_VERSION >= 9000
if (*comm->intraCC && (ncclCudaFullCompCap() == *comm->intraCC)) {
// Check whether the GPU supports Cooperative Group Multi Device Launch
(void) cudaDeviceGetAttribute(&cgMdLaunch, cudaDevAttrCooperativeMultiDeviceLaunch, comm->cudaDev);