I have added a new mode:
this supplements the previous two modes:
--perfcounter-model process --perfcounter-model core
This mode works similar to the process model, except only the root process is instrumented and then the “inherit” flag is passed. This lets me compare the values of inherit vs. totaling all the counters of a tree.
In principle, this is also compatible with the ftrace processtree model, though I haven’t turned that plumbing on yet.