This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug gcov-profile/71672] inlining indirect calls does not work with autofdo


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71672

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW

--- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
Good, using perf 4.6 works for me. However the profile is rejected here:

  if (total >= info->count / 2)
    {
      if (dump_file)
        fprintf (dump_file, " not hot anymore %ld >= %ld",
                 (long)total,
                 (long)info->count /2);
      return false;
    }

where total == 388 and info->count == 360. IIUC the count belongs to a BB where
the indirect call is triggered? If so, the bail out does not make much sense.
Anyhow, commenting out the return statement, I fails dues to "no target map".

Debugging find_icall_target_map function which is correctly called for p_10 (),
but callsites of main is just single and equal to 'setp'. That's strange, as
the call is completely inlined.

Simple dump patch generates:

f: a1, num_callsites: 0
num_targets: 0
f: main, num_callsites: 1
num_targets: 0
num_targets: 0
num_targets: 1
num_targets: 0
f: setp, num_callsites: 0
num_targets: 0
num_targets: 0

So it looks main contains just the callsite to 'setp', call of 'p' is missing.
Would it be related to a gcov_create issue. Any verbose debug option I can use?

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]