Re: Perf stat regression from d15480a3d67

[Andi Kleen <ak@linux.intel.com>]

> > Okay so the skipping can just be removed?
> 
> The existing use case was system-wide --per-thread mode.
> It would generate too much noise without it.

Hmm I guess we could make it default for per thread mode, but add an
option to turn it on again if it's needed

(my tool currently doesn't do per thread, but I can could imagine
adding that at some point)

> > > Didn't you say you only have problems in uncore events?
> >
> > For this test case I only hit it with the uncore, but I would have a general
> > problem for any event if it was hidden.
> 
> stat-display.c::should_skip_zero_counter() specifically checks
> uncore PMU events, so I don't think core events have problems.
> 
> Also it only skips when aggregation id doesn't match, which means
> the (default) global aggregation mode should not skip any uncore
> events.
> 
> Can you please help me to find a minimal repro?

I'll run delta[1] over it later today.
But you may need a SKX.

-Andi

[1] https://github.com/dsw/delta