Closed Palwisha-18 closed 4 years ago
Hey @Palwisha-18,
Comm|Scope uses the Google Benchmark library to control the actual benchmarking runs, and their position seems to be that a user usually doesn't actually need to control the number of iterations, so they don't expose it as a command line option.
However, you should be able to modify the benchmark you want to run with this API where the benchmark is registered, for example, something like:
benchmark::RegisterBenchmark(...)->Iterations(10);
I haven't tried this myself, so let me know if it doesn't work.
For further discussion, see google/benchmark#370, 373, google/benchmark#370.
It worked! Thanks!
Excellent! I saw https://github.com/Palwisha-18/ComScribe, looks like a worthwhile project. That is an area of interest to me, so let me know if you have any more questions.
I also think your colleague @erhant might be interested in https://github.com/cwpearson/stencil for his 3D stencil GPU communication work. He may also want to look at the related paper https://cwpearson.github.io/publication/2020522_pearson_iwapt
Cheers!
Thank you Carl for your feedback on ComScribe. We have a paper in submission for this project and we would like to share it with you when it gets accepted. Thanks :)
Also thank you very much for your suggestions on my topic! I will be looking into them.
Hi,
After running any micro-benchmark several times, it is observed that the number of iterations varies. PFA the log for HostToGPU memcpy to what iteration term I am referring to. Is it possible to fix the iterations through the command line?
Benchmark Time CPU Iterations UserCounters...
Comm_Memcpy_HostToGPU/log2(N):8/manual_time 10165 ns 122345 ns 76502 bytes=256 bytes_per_second=24.0178M/s cuda_id=0