[Bug gcov-profile/58127] New: [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[Bug gcov-profile/58127] New: [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

            Bug ID: 58127
           Summary: [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for
                    x86_64-apple-darwin10
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: gcov-profile
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dominiq at lps dot ens.fr
                CC: hubicka at gcc dot gnu.org, iains at gcc dot gnu.org
              Host: x86_64-apple-darwin10
            Target: x86_64-apple-darwin10
             Build: x86_64-apple-darwin10

Between revisions 201611 (OK) and 201635 the following failures appeared:

FAIL: gcc.dg/tree-prof/20041218-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/bb-reorg.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/comp-goto-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/crossmodule-indircall-1a.c compilation, 
-fprofile-generate -D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/ic-misattribution-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/ic-misattribution-1a.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/indir-call-prof.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/inliner-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/peel-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr34999.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr44777.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr45354.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr47187.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr49299-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr49299-2.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr50907.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr52027.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/pr52150.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/prof-robust-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/stringop-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/stringop-2.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/switch-case-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/switch-case-2.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/tracer-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/unroll-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/update-cunroll-2.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/update-loopch.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/update-tailcall.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/va-arg-pack-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/val-prof-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/val-prof-2.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/val-prof-3.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/val-prof-4.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/val-prof-5.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/val-prof-6.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/val-prof-7.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
FAIL: gcc.dg/tree-prof/wcoverage-mismatch.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE

The errors are

output is:
Undefined symbols:
  "___gcov_indirect_call_callee", referenced from:
      ___gcov_indirect_call_callee$non_lazy_ptr in ccNJUvcM.o
     (maybe you meant: ___gcov_indirect_call_callee$non_lazy_ptr)
ld: symbol(s) not found
collect2: error: ld returned 1 exit status

FAIL: gcc.dg/tree-prof/20041218-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
UNRESOLVED: gcc.dg/tree-prof/20041218-1.c execution,    -fprofile-generate
-D_PROFILE_GENERATE
UNRESOLVED: gcc.dg/tree-prof/20041218-1.c compilation,  -fprofile-use
-D_PROFILE_USE
UNRESOLVED: gcc.dg/tree-prof/20041218-1.c execution,    -fprofile-use
-D_PROFILE_USE

or

output is:
Undefined symbols:^M
  "___gcov_indirect_call_callee", referenced from:^M
      ___gcov_indirect_call_callee$non_lazy_ptr in ccLKIDNc.o^M
      ___gcov_indirect_call_callee$non_lazy_ptr in ccLrBSua.o^M
     (maybe you meant: ___gcov_indirect_call_callee$non_lazy_ptr)^M
  "___gcov_indirect_call_counters", referenced from:^M
      ___gcov_indirect_call_counters$non_lazy_ptr in ccLKIDNc.o^M
      ___gcov_indirect_call_counters$non_lazy_ptr in ccLrBSua.o^M
     (maybe you meant: ___gcov_indirect_call_counters$non_lazy_ptr)^M
ld: symbol(s) not found^M
collect2: error: ld returned 1 exit status^M

FAIL: gcc.dg/tree-prof/ic-misattribution-1.c compilation,  -fprofile-generate
-D_PROFILE_GENERATE
UNRESOLVED: gcc.dg/tree-prof/ic-misattribution-1.c execution,   
-fprofile-generate -D_PROFILE_GENERATE
UNRESOLVED: gcc.dg/tree-prof/ic-misattribution-1.c compilation,  -fprofile-use
-D_PROFILE_USE
UNRESOLVED: gcc.dg/tree-prof/ic-misattribution-1.c execution,    -fprofile-use
-D_PROFILE_USE

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #1 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Revision 201632 is OK, r201634 is not.

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-08-17
     Ever confirmed|0                           |1

--- Comment #2 from Iain Sandoe <iains at gcc dot gnu.org> ---
confirmed on x86_64-darwin10/12.
possibly something needs to be updated in the emutls implementation?

$ nm -g gcc/libgcov.a |grep gcov_indirect_call_callee
00000000000000a0 D ___emutls_v.__gcov_indirect_call_callee

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #3 from Iain Sandoe <iains at gcc dot gnu.org> ---
(In reply to Iain Sandoe from comment #2)
> confirmed on x86_64-darwin10/12.
> possibly something needs to be updated in the emutls implementation?
> 
> $ nm -g gcc/libgcov.a |grep gcov_indirect_call_callee
> 00000000000000a0 D ___emutls_v.__gcov_indirect_call_callee

.. at least, 
pass_ipa_profile is running _after_ pass_ipa_lower_emutls.

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED

--- Comment #4 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
The secret plan was to not use TLS for emutls. I did not introduce new use of
TLS, just use what was already in tree-profile.c:
  if (targetm.have_tls)
    DECL_TLS_MODEL (ic_void_ptr_var) =
      decl_default_tls_model (ic_void_ptr_var);

and I think have_tls should be true only for non-emutls targets.
on libgcov time we use
#ifdef HAVE_CC_TLS
__thread
#endif

There seems to be macro USE_EMUTLS specifying if we have real TLS or emuTLS.

I am not sure what of the two tests passes.  Perhaps it will help to replace
#ifdef HAVE_CC_TLS
__thread
#endif
#if defined(HAVE_CC_TLS) && !defined (USE_EMUTLS)
__thread
#endif

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #5 from Iain Sandoe <iains at gcc dot gnu.org> ---
(In reply to Jan Hubicka from comment #4)
> The secret plan was to not use TLS for emutls.

OK - I'll take a look at what should be used for the test...

> There seems to be macro USE_EMUTLS specifying if we have real TLS or emuTLS.
> 
> I am not sure what of the two tests passes.  Perhaps it will help to replace
> #ifdef HAVE_CC_TLS
> __thread
> #endif
> #if defined(HAVE_CC_TLS) && !defined (USE_EMUTLS)
> __thread
> #endif

... seems plausible, will try it on my next build cycle.

BTW, I wonder what it would take to make it work with emutls?

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #6 from Iain Sandoe <iains at gcc dot gnu.org> ---
(In reply to Iain Sandoe from comment #5)
> (In reply to Jan Hubicka from comment #4)
> > The secret plan was to not use TLS for emutls.

that's kind of a pity, since IIRC one of the main reasons for recasting the
emutls implementation as a separate pass was to allow for uses like this; 
but, having said that, probably not a high priority.

> > There seems to be macro USE_EMUTLS specifying if we have real TLS or emuTLS.

<snip>

> ... seems plausible, will try it on my next build cycle.

#if defined(HAVE_CC_TLS) && !defined (USE_EMUTLS)

does, indeed, work.

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[hubicka at ucw dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #7 from Jan Hubicka <hubicka at ucw dot cz> ---
> 
> that's kind of a pity, since IIRC one of the main reasons for recasting the
> emutls implementation as a separate pass was to allow for uses like this; 
> but, having said that, probably not a high priority.
MThere are performance implications of doing so.   One way would be to add
facility to insert new variables into emutls late (it is probably doable).
Other easy way around would be to add gcov function to set both pointers to
proper values and do it as function call.  

I am not sure if this is feasible performance wise - the counter is in TLS
just to avoid too many colisions. Generally profiling should work resonably
well even without TLS for that.

> 
> #if defined(HAVE_CC_TLS) && !defined (USE_EMUTLS)
> 
> does, indeed, work.

If you do have chance to commit it, consider the patch preapproved.  I 
will be offline during the afternoon.

Honza

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #8 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> 
> #if defined(HAVE_CC_TLS) && !defined (USE_EMUTLS)
> 
> does, indeed, work.

I have done a clean bootstrap with

--- ../_clean/libgcc/libgcov.c    2013-08-11 13:28:22.000000000 +0200
+++ libgcc/libgcov.c    2013-08-18 12:30:01.000000000 +0200
@@ -1162,11 +1162,11 @@ __gcov_indirect_call_profiler (gcov_type
    The variables are set directly by GCC instrumented code, so declaration
    here must match one in tree-profile.c  */

-#ifdef HAVE_CC_TLS
+#if defined(HAVE_CC_TLS) && !defined (USE_EMUTLS)
 __thread 
 #endif
 void * __gcov_indirect_call_callee;
-#ifdef HAVE_CC_TLS
+#if defined(HAVE_CC_TLS) && !defined (USE_EMUTLS)
 __thread 
 #endif
 gcov_type * __gcov_indirect_call_counters;

and the failures for gcc.dg/tree-prof/* and g++.dg/tree-prof/* are gone.
Thanks.

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

Iain Sandoe <iains at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #10 from Iain Sandoe <iains at gcc dot gnu.org> ---
thus fixed.

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[iains at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #9 from Iain Sandoe <iains at gcc dot gnu.org> ---
(In reply to Jan Hubicka from comment #7)
> > 
> > that's kind of a pity, since IIRC one of the main reasons for recasting the
> > emutls implementation as a separate pass was to allow for uses like this; 
> > but, having said that, probably not a high priority.
> MThere are performance implications of doing so.   One way would be to add
> facility to insert new variables into emutls late (it is probably doable).
> Other easy way around would be to add gcov function to set both pointers to
> proper values and do it as function call.  
> 
> I am not sure if this is feasible performance wise - the counter is in TLS
> just to avoid too many colisions. Generally profiling should work resonably
> well even without TLS for that.

ack, let's not pursue that then.

> If you do have chance to commit it, consider the patch preapproved.  

applied as 201829

libgcc:

    PR gcov-profile/58127
    * libgcov.c (__gcov_indirect_call_callee): Don't make this a
    __thread var for emulated TLS.
    (__gcov_indirect_call_counters): Likewise.

[Bug gcov-profile/58127] [4.9 Regression] 37 failures in gcc.dg/tree-prof/ for x86_64-apple-darwin10

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58127

--- Comment #11 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> thus fixed.

Thanks!