Re: [PATCH 4/4] perf, tools: Add perf stat --transaction v3

[Arnaldo Carvalho de Melo <acme@ghostprotocols.net>]

Em Thu, Aug 15, 2013 at 04:29:25PM +0200, Andi Kleen escreveu:
> > >   * Update various tracking values we maintain to print
> > >   * more semantic information such as miss/hit ratios,
> > > @@ -283,8 +340,12 @@ static void update_shadow_stats(struct perf_evsel *counter, u64 *count)
> > >  		update_stats(&runtime_nsecs_stats[0], count[0]);
> > >  	else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES))
> > >  		update_stats(&runtime_cycles_stats[0], count[0]);
> > > -	else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND))
> > > -		update_stats(&runtime_stalled_cycles_front_stats[0], count[0]);
> > 
> > Why remove the test for HW_STALLED_CYCLES_FRONTEND?
> 
> Hmm that was probably a merge error.
> 
> I'll fix & resend.

Ok
 
> > Got it why it doesn't need to account for the '{' in the array ;-)
> > 
> > While this works and isn't in any fast path, I find it ugly with all
> > this looping in nth_evsel.
> > 
> > Why not:
> > 
> > 	} else if (evsel->idx == T_CYCLES_IN_TX)) &&
> > 
> > ? I guess this works as you expect, no?
> 
> I had some problems with people using -T, but also setting custom
> events, that is why I added the extra comparison

But it'd fail the same, as if you go from the first entry in the
evlist->entries linked list, looping while incrementing a number, that
number should match the nth entry ->idx member.

I'll try to experiment with this when you resend with the parts you
agreed should be changed.
 
> The event lists are small enough that it's not really noticeable.
> I suppose could set up an array once.

Its not about performance hits, they are negligible, but about using
something that should provide equivalent results.

- Arnaldo