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

[TransferBench] Adding more preset benchmarks to filter read mode, cpu vs gpu pairs (#477)

Browse Source 
[ROCm/rccl commit: 1157c2edfe]
This commit is contained in:
gilbertlee-amd
2021-11-24 18:05:37 -07:00
committed by GitHub
parent 2da8358a74
commit ef1cbb03d2
2 changed files with 53 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/tools/TransferBench/TransferBench.hpp
+1 -1
View File
@@ -122,7 +122,7 @@ void DeallocateMemory(MemType memType, int devIndex, float* memPtr);
void CheckPages(char* byteArray, size_t numBytes, int targetId);
void CheckOrFill(ModeType mode, int N, bool isMemset, bool isHipCall, std::vector<float> const& fillPattern, float* ptr);
void RunLink(EnvVars const& ev, size_t const N, int const iteration, Link& link);
void RunPeerToPeerBenchmarks(EnvVars const& ev, size_t N, int numBlocksToUse);
void RunPeerToPeerBenchmarks(EnvVars const& ev, size_t N, int numBlocksToUse, int readMode, int skipCpu);
double GetPeakBandwidth(EnvVars const& ev, size_t N, int isBidirectional,
MemType srcMemType, int srcIndex,
MemType dstMemType, int dstIndex,