Enable collecting metrics from a set

[arjunmenonv]

Can you extend the metric argument in nsight.analyze.kernel to support collecting all metrics from a set? Manually picking the metrics for each section in the set is cumbersome. I understand that this slows the benchmarking down and results in bloated output csv files, but would like to have this as an option when I need detailed ncu reports.

[nvssatoor]

@arjunmenonv would you want to collect the metrics from a set for different kernel configurations and multiple runs for each configuration as supported for a single metric?
Or this would be used only for a single configuration and single run?

[arjunmenonv]

Hi @nvssatoor, sorry I missed your reply.
Ideally I'd like to collect all the metrics for a single kernel configuration with a single benchmark run, like we do with the Nsight-compute cli tool. Support for different kernel configurations is nice to have, but I am ok writing some python wrappers / bash scripts to do this at my end.

[nvssatoor]

@arjunmenonv we are evaluating enhancing nsight python to support more metric collection options. This should also cover your request.

Thanks