Re: [GIT PULL 00/21] perf/core improvements and fixes

From: Arnaldo Carvalho de Melo <acme@infradead.org>
To: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>,
	Ingo Molnar <mingo@kernel.org>,
	linux-kernel@vger.kernel.org, Andi Kleen <ak@linux.intel.com>,
	Corey Ashford <cjashfor@linux.vnet.ibm.com>,
	David Ahern <dsahern@gmail.com>,
	Frederic Weisbecker <fweisbec@gmail.com>,
	Irina Tirdea <irina.tirdea@gmail.com>,
	Mike Galbraith <efault@gmx.de>, Paul Mackerras <paulus@samba.org>,
	Peter Zijlstra <peterz@infradead.org>,
	stable@vger.kernel.org, Stephane Eranian <eranian@google.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Vinson Lee <vlee@twitter.com>, Zheng Liu <wenqing.lz@taobao.com>
Subject: Re: [GIT PULL 00/21] perf/core improvements and fixes
Date: Mon, 12 Nov 2012 13:01:39 -0300	[thread overview]
Message-ID: <20121112160139.GC18978@ghostprotocols.net> (raw)
In-Reply-To: <20121112135545.GA4595@krava.brq.redhat.com>

Em Mon, Nov 12, 2012 at 02:55:46PM +0100, Jiri Olsa escreveu:
> On Mon, Nov 12, 2012 at 11:10:52AM +0900, Namhyung Kim wrote:
> > On Fri,  9 Nov 2012 18:42:49 -0300, Arnaldo Carvalho de Melo wrote:
> > > . Add a 'link' method for hists, so that we can have the leader with
> > >   buckets for all the entries in all the hists.  This new method
> > >   is now used in the default 'diff' output, making the sum of the 'baseline'
> > >   column be 100%, eliminating blind spots. Now we need to use this
> > >   for 'diff' with > 2 perf.data files and for multi event 'report' and
> > >   'annotate'.

> > I'm not sure it can be used for group report at least in its current
> > form.  IIUC it connects multiple hist entries using a list head and
> > create a dummy entry in the leader if need be.  But it didn't handle
> > non-leader entries so it's hard to tell which is which if less entries
> > are present only.  For example consider following case:

> > 	leader		member1		member2
> > 	A		A		A
> > 	B
> > 			C
> > 					D

> > where leader, member1 and member2 are evsel/hists and A, B, C and D are
> > hist entries.  After 'linking' the entries the leader will have
> > following linkage:

> > 	leader
> > 	A	->	A	->	A
> > 	B
> > 	C (dummy) ->	C
> > 	D (dummy)		->	D

> > In this case, for entry A the leader can determine which entry came from
> > which hists by looking its order in the list.  For entry B the leader
> > can use zero value for them since the list is empty.  However for
> > entries C and D, it cannot know which one is the right hists unless it
> > records a hist index or creates dummy entry and insert it in a correct
> > order (looks far from an optimal solution).  Am I missing something?

> there's hists pointer in hist_entry if that's what you look for

And from there to evsel->idx. In your patchset you even introduce
hists_2_evsel(), right?

- Arnaldo