Re: Implement lbr-as-callgraph v9

[Andi Kleen <andi@firstfloor.org>]

On Tue, Aug 12, 2014 at 03:50:17PM +0200, Jiri Olsa wrote:
> On Thu, Jul 31, 2014 at 10:17:00AM -0700, Andi Kleen wrote:
> 
> SNIP
> 
> > v5:
> > - Rename functions
> > - Fix gtk build problem
> > - Fix crash without -g
> > - Improve error messages
> > - Improve srcline display in various ways
> > v6:
> > - Port to latest perf/core
> > v7:
> > - Really port to latest perf/core
> > v8:
> > - Rebased on 3.16-rc1
> > v9: 
> > - Forward ported to latest tip/perf/core
> > 
> > Example output:
> > 
> >     % perf record -b -g ./tsrc/tcall
> 
> Why do we need the '-g' in here.. '-b' should be enough right?
> I saw u fill the rest of the backtrace with callchain
> data (if needed), but thats not necessary right?

Right now the callgraph processing needs -g to enable itself.
I didn't try to fix it so far because we likely use -g in these
situations where --branch-history is useful.

If you think it's important it could be fixed I think.

> 
> 
> >     [ perf record: Woken up 1 times to write data ]
> >     [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ]
> >     % perf report --branch-history
> 
> When I captured data like you above and run pure 'perf report'
> I did not get callchains displayed at all.
> 
> Also using 'perf report --branch-history' enables filename:line
> suffix for callchain regardless of using '-b' for data capture

That's fine isn't it? If the user specifies that option
they should have used -b.

Or do you mean it should only do it for branch entries?
Right now the callgraph code doesn't distinguish between branch
entry and callgraph entry, so it can't know. 

-Andi