[PATCH] gcov: Mark BBs that do not correspond to a line in source code (PR gcov-profile/79891).

[Richard Biener]

On Tue, 14 Mar 2017, Richard Biener wrote:

> On Tue, 14 Mar 2017, Martin Liška wrote:
> > 
> > /tmp/gcov-1.gcno:	block 2:`/home/marxin/Programming/gcc/gcc/testsuite/gcc.misc-tests/gcov-1.c':10, 14
> > /tmp/gcov-1.gcno:	block 4:`/home/marxin/Programming/gcc/gcc/testsuite/gcc.misc-tests/gcov-1.c':14
> > 
> > where blocks 2 and 4 are:
> > 
> >   <bb 2> [0.00%]:
> >   i_3 = 0;
> >   goto <bb 4>; [0.00%]
> > 
> > ...
> > 
> >   <bb 4> [0.00%]:
> >   # i_1 = PHI <i_3(2), i_7(3)>
> >   if (i_1 <= 9)
> >     goto <bb 3>; [0.00%]
> >   else
> >     goto <bb 5>; [0.00%]
> > 
> > The same happens to int a = b < 1 ? (c < 3 ? d : c) : a;
> > 
> > /tmp/gcov2.gcno:	block 2:`/tmp/gcov2.c':1, 3
> > 
> >   <bb 2> [0.00%]:
> >   if (b_3(D) <= 0)
> >     goto <bb 3>; [0.00%]
> >   else
> >     goto <bb 7>; [0.00%]
> > 
> > That showed a caching of locations actually magically handles loops and ternary operations.
> > I'm still wondering how should be defined line count for a multiple statements happening
> > on the line? Having that we can find a proper solution.
> 
> It should be number of times the line is _entered_, that is, lineno
> changed from something != lineno to lineno.  Consider
> 
> foo (); goto baz; lab: bar ();   // line 1
> baz:
> goto lab;
> 
> should increment line 1 when entering to foo () as well as when
> entering through goto lab.  but both times just once.

Of course with -a you are supposed to get sub-line accuracy for these
cases, but then you'll get not a single number per line (and not sure
how you can reasonably interpret -a info without column info or so).

Richard.

> Richard.
> 
> 
> > Martin
> > 
> > > 
> > >           for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
> > >             {
> > >               gimple *stmt = gsi_stmt (gsi);
> > >               if (!RESERVED_LOCATION_P (gimple_location (stmt)))
> > >                 output_location (gimple_filename (stmt), gimple_lineno 
> > > (stmt),
> > >                                  &offset, bb);
> > > 
> > > should use expand_location and then look at the spelling location,
> > > otherwise we'll get interesting effects with macro expansion?
> > > 
> > >             }
> > > 
> > > Richard.
> > > 
> > >> Martin
> > >>
> > >>
> > >>>
> > >>> Richard.
> > >>>
> > >>>> Martin
> > >>>>
> > >>>>>
> > >>>>>> Martin
> > >>>>>>
> > >>>>>>>
> > >>>>>>>
> > >>>>>>>> Martin
> > >>>>>>>>
> > >>>>>>>>>
> > >>>>>>>>> Richard.
> > >>>>>>>>>
> > >>>>>>>>>
> > >>>>>>>>>> Hope Nathan will find time to provide review as he's familiar with content of gcov.c.
> > >>>>>>>>>>
> > >>>>>>>>>> Martin
> > >>>>>>>>>>
> > >>>>>>>>>>>
> > >>>>>>>>>>> OTOH I don't know much about gcov format.
> > >>>>>>>>>>>
> > >>>>>>>>>>> Richard.
> > >>>>>>>>>>>
> > >>>>>>>>>>>> Martin
> > >>>>>>>>>>>>
> > >>>>>>>>>>>>>
> > >>>>>>>>>>>>> Richard.
> > >>>>>>>>>>>>>
> > >>>>>>>>>>>>>> Martin
> > >>>>>>>>>>>>>>
> > >>>>>>>>>>>>>>>
> > >>>>>>>>>>>>>>> Richard.
> > >>>>>>>>>>>>>>>
> > >>>>>>>>>>>>>>
> > >>>>>>>>>>>>>>
> > >>>>>>>>>>>>>
> > >>>>>>>>>>>>
> > >>>>>>>>>>>>
> > >>>>>>>>>>>
> > >>>>>>>>>>
> > >>>>>>>>>>
> > >>>>>>>>>
> > >>>>>>>>
> > >>>>>>>>
> > >>>>>>>
> > >>>>>>
> > >>>>>>
> > >>>>>
> > >>>>
> > >>>>
> > >>>
> > >>
> > >>
> > > 
> > 
> > 
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nuernberg)