From mboxrd@z Thu Jan  1 00:00:00 1970
From: Arun Sharma <asharma@fb.com>
Subject: Re: [PATCH 2/2] perf: Add a new sort order: SORT_INCLUSIVE
Date: Thu, 8 Mar 2012 10:22:26 -0800
Message-ID: <20120308182226.GA10065@dev3310.snc6.facebook.com>
References: <1331160079-13821-1-git-send-email-asharma@fb.com>
 <1331160079-13821-3-git-send-email-asharma@fb.com>
 <20120308153933.GD7976@somewhere.redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Return-path: <linux-perf-users-owner@vger.kernel.org>
Received: from outmail024.snc4.facebook.com ([66.220.144.158]:59780 "EHLO
	intmgw005.snc4.facebook.com" rhost-flags-OK-OK-OK-FAIL)
	by vger.kernel.org with ESMTP id S1755598Ab2CHSWb (ORCPT
	<rfc822;linux-perf-users@vger.kernel.org>);
	Thu, 8 Mar 2012 13:22:31 -0500
Content-Disposition: inline
In-Reply-To: <20120308153933.GD7976@somewhere.redhat.com>
Sender: linux-perf-users-owner@vger.kernel.org
List-ID: <linux-perf-users.vger.kernel.org>
To: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arun Sharma <asharma@fb.com>, linux-kernel@vger.kernel.org, Ingo Molnar <mingo@elte.hu>, Arnaldo Carvalho de Melo <acme@redhat.com>, Mike Galbraith <efault@gmx.de>, Paul Mackerras <paulus@samba.org>, Peter Zijlstra <peterz@infradead.org>, Stephane Eranian <eranian@google.com>, Namhyung Kim <namhyung.kim@lge.com>, Tom Zanussi <tzanussi@gmail.com>, linux-perf-users@vger.kernel.org

On Thu, Mar 08, 2012 at 04:39:36PM +0100, Frederic Weisbecker wrote:
> 
> I don't yet understand the point of this.
> 
> Imagine those three hists:
> 
> a -> b -> c
> a -> b -> d
> a-> e -> f
> 
> The fractal inverted mode (-G) will report this:
> 
> a--
>   |
>   ----- b
>   |     |
>   |     -----c
>   |     |
>   |     -----d
>   |
>   ----- e
>         |
>         -----f
> 

Please see the test program attached. It has only two paths from main()
to c(). But for this discussion, imagine a callgraph with 10 different
paths.

With -G, c() appears 10 times in the callgraph and the user is required
to manually sum up the samples to realize that the callgraph under c()
is very expensive.

With -s inclusive, c() will show up at the very top after main().

 -Arun

#include <stdio.h>

int sum = 0;

#define LOOP(n)                                         \
        {                                               \
                int j;                                  \
                for (j = 0; j < 10000; j++) {           \
                        sum += j;                       \
                }                                       \
        }

int f()
{
        LOOP(100);
}

int d()
{
        LOOP(100);
        f();
}

int e()
{
	LOOP(100);
	f();
}

int c()
{
	LOOP(100);
	d();
	LOOP(100);
	e();
}

int b() 
{
	LOOP(70);
	c();
}

int a() 
{
	LOOP(30);
	c();
}

int main()
{
	int i;
	for (i = 0; i < 10000; i++) {
		a();
		b();
	}
}