RE: our internal debug, we noticed counters from a pmc: line in a file, occasionally get split over multiple runs in pmcperf<#>.txt. This results in nonsensical values sometimes, due to-to-run variation.
Need to debug how this could happen.
Seems like some sort of non-deterministic ordering in a python API / pandas / etc.
https://github.com/ROCm/omniperf/blob/59254a50463a8f3fe44c6b1f06610edf53e2c30c/src/omniperf_profile/profiler_base.py#L90
RE: our internal debug, we noticed counters from a
pmc:
line in a file, occasionally get split over multiple runs in pmcperf<#>.txt. This results in nonsensical values sometimes, due to-to-run variation.Need to debug how this could happen.
Seems like some sort of non-deterministic ordering in a python API / pandas / etc.