Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 10/13/2016 04:31 PM, Rainer Orth wrote:
> Hi Martin,
> 
>> Good! How does it look with the former solaris targets that does not support
>> prioritized ctors?
> 
> still no failures there, neither with ld (which lacks constructor
> priority support) nor with gld (which has it).  Only Solaris 12 shows
> the failures, both with ld and gld (both of which support constructor
> priorities).
> 
>   Rainer
> 

I see. So please send me some example of a binary that still fails
on Solaris 12.

Thanks,
Martin

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

> Good! How does it look with the former solaris targets that does not support
> prioritized ctors?

still no failures there, neither with ld (which lacks constructor
priority support) nor with gld (which has it).  Only Solaris 12 shows
the failures, both with ld and gld (both of which support constructor
priorities).

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 10/13/2016 04:04 PM, Rainer Orth wrote:
> no, it's from r240990 unlike I'm completely mistaken.  However, current
> trunk bootstraps are running as we speak.
> 
>   Rainer

Good! How does it look with the former solaris targets that does not support
prioritized ctors?

Thanks,
Martin

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

> Just running my previous example (priotity.c), I can see with -S:
>
> _GLOBAL__sub_D_00099_1_foo:
>   jmp __gcov_exit
>   .size   _GLOBAL__sub_D_00099_1_foo, .-_GLOBAL__sub_D_00099_1_foo
>   .section.fini_array.00099,"aw"
>   .align 4
>   .long   _GLOBAL__sub_D_00099_1_foo
>   .data
>   .align 4
>   .type   .LPBX1, @object
>   .size   .LPBX1, 12
>
> Which looks good. I guess the sent snippet is before r240857, where I fixed 
> the
> priority to 99. Can you please test it with current trunk?

no, it's from r240990 unlike I'm completely mistaken.  However, current
trunk bootstraps are running as we speak.

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 10/13/2016 03:46 PM, Rainer Orth wrote:
> Hi Martin,
> 
> sorry for the long delay: I've been extremely busy the last two weeks.

Hello

Never mind, still plenty of time before we'll release 7.1.0 :)

> 
>> On 09/30/2016 02:31 PM, Rainer Orth wrote:
>>> this would be i386-pc-solaris2.12.  I'm not sure if the constructor
>>> priority detection works in a cross scenario.
>>
>> Hi.
>>
>> By the way, I tried to test the cross-compiler:
>> $ ../configure --disable-bootstrap --enable-languages=c,c++,fortran 
>> --enable-valgrind-annotations --prefix=/home/marxin/bin/gcc2 
>> --disable-multilib --disable-libsanitizer --target=i386-pc-solaris2.12
>>
>> and I get for:
>> cat /tmp/priority.c
>> void __attribute__ ((constructor(150))) foo()
>> {
>> }
>>
>> void __attribute__ ((constructor(151))) bar()
>> {
>> }
>>
>> int main()
>> {
>>   return 0;
>> }
>>
>> $ ./xgcc -B. /tmp/priority.c -fprofile-generate -S
>> /tmp/priority.c:2:1: error: constructor priorities are not supported
>>  {
>>  ^
>> /tmp/priority.c:6:1: error: constructor priorities are not supported
>>  {
>>  ^
>>
>> I guess even cross compiler should detect whether the target supports 
>> ctor/dtor priorities.
> 
> maybe it could, but right now acinclude.m4 (gcc_AC_INITFINI_ARRAY) has
> this for crosses:
> 
> AC_MSG_CHECKING(cross compile... guessing)
> gcc_cv_initfini_array=no
> 
> You could work around this by overriding configure with
> --enable-initfini-array.

Good, I've just done that.

> 
>> May I ask you for assembly file of a native compiler with the suggested 
>> patch?
> 
> Sure: this time from an i386-pc-solaris2.12 compiler configured to use
> gas and ld.
> 
>   Rainer
> 

Just running my previous example (priotity.c), I can see with -S:

_GLOBAL__sub_D_00099_1_foo:
jmp __gcov_exit
.size   _GLOBAL__sub_D_00099_1_foo, .-_GLOBAL__sub_D_00099_1_foo
.section.fini_array.00099,"aw"
.align 4
.long   _GLOBAL__sub_D_00099_1_foo
.data
.align 4
.type   .LPBX1, @object
.size   .LPBX1, 12

Which looks good. I guess the sent snippet is before r240857, where I fixed the
priority to 99. Can you please test it with current trunk?

Thanks,
Martin

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

sorry for the long delay: I've been extremely busy the last two weeks.

> On 09/30/2016 02:31 PM, Rainer Orth wrote:
>> this would be i386-pc-solaris2.12.  I'm not sure if the constructor
>> priority detection works in a cross scenario.
>
> Hi.
>
> By the way, I tried to test the cross-compiler:
> $ ../configure --disable-bootstrap --enable-languages=c,c++,fortran 
> --enable-valgrind-annotations --prefix=/home/marxin/bin/gcc2 
> --disable-multilib --disable-libsanitizer --target=i386-pc-solaris2.12
>
> and I get for:
> cat /tmp/priority.c
> void __attribute__ ((constructor(150))) foo()
> {
> }
>
> void __attribute__ ((constructor(151))) bar()
> {
> }
>
> int main()
> {
>   return 0;
> }
>
> $ ./xgcc -B. /tmp/priority.c -fprofile-generate -S
> /tmp/priority.c:2:1: error: constructor priorities are not supported
>  {
>  ^
> /tmp/priority.c:6:1: error: constructor priorities are not supported
>  {
>  ^
>
> I guess even cross compiler should detect whether the target supports 
> ctor/dtor priorities.

maybe it could, but right now acinclude.m4 (gcc_AC_INITFINI_ARRAY) has
this for crosses:

AC_MSG_CHECKING(cross compile... guessing)
gcc_cv_initfini_array=no

You could work around this by overriding configure with
--enable-initfini-array.

> May I ask you for assembly file of a native compiler with the suggested patch?

Sure: this time from an i386-pc-solaris2.12 compiler configured to use
gas and ld.

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University


.file   "pr16855.C"
.globl  a
.bss
.align 4
.type   a, @object
.size   a, 4
a:
.zero   4
.text
.globl  _Z3foov
.type   _Z3foov, @function
_Z3foov:
.LFB55:
.cfi_startproc
pushl   %ebp
.cfi_def_cfa_offset 8
.cfi_offset 5, -8
movl%esp, %ebp
.cfi_def_cfa_register 5
movl__gcov0._Z3foov, %eax
movl__gcov0._Z3foov+4, %edx
addl$1, %eax
adcl$0, %edx
movl%eax, __gcov0._Z3foov
movl%edx, __gcov0._Z3foov+4
movl$123, a
nop
popl%ebp
.cfi_restore 5
.cfi_def_cfa 4, 4
ret
.cfi_endproc
.LFE55:
.size   _Z3foov, .-_Z3foov
.data
.align 4
.type   _ZL16__gthread_active, @object
.size   _ZL16__gthread_active, 4
_ZL16__gthread_active:
.long   -1
.text
.type   _ZL17__gthread_triggerv, @function
_ZL17__gthread_triggerv:
.LFB336:
.cfi_startproc
pushl   %ebp
.cfi_def_cfa_offset 8
.cfi_offset 5, -8
movl%esp, %ebp
.cfi_def_cfa_register 5
movl__gcov0._ZL17__gthread_triggerv, %eax
movl__gcov0._ZL17__gthread_triggerv+4, %edx
addl$1, %eax
adcl$0, %edx
movl%eax, __gcov0._ZL17__gthread_triggerv
movl%edx, __gcov0._ZL17__gthread_triggerv+4
movl$1, _ZL16__gthread_active
nop
popl%ebp
.cfi_restore 5
.cfi_def_cfa 4, 4
ret
.cfi_endproc
.LFE336:
.size   _ZL17__gthread_triggerv, .-_ZL17__gthread_triggerv
.local  _ZStL8__ioinit
.comm   _ZStL8__ioinit,1,1
.section.rodata
.LC0:
.string "In Test ctor"
.section.text._ZN4TestC2Ev,"axG",@progbits,_ZN4TestC5Ev,comdat
.align 2
.weak   _ZN4TestC2Ev
.type   _ZN4TestC2Ev, @function
_ZN4TestC2Ev:
.LFB1086:
.cfi_startproc
pushl   %ebp
.cfi_def_cfa_offset 8
.cfi_offset 5, -8
movl%esp, %ebp
.cfi_def_cfa_register 5
subl$8, %esp
movl__gcov0._ZN4TestC2Ev, %eax
movl__gcov0._ZN4TestC2Ev+4, %edx
addl$1, %eax
adcl$0, %edx
movl%eax, __gcov0._ZN4TestC2Ev
movl%edx, __gcov0._ZN4TestC2Ev+4
subl$8, %esp
pushl   $.LC0
pushl   $_ZSt4cout
call_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_PKc
addl$16, %esp
movl%eax, %ecx
movl__gcov0._ZN4TestC2Ev+8, %eax
movl__gcov0._ZN4TestC2Ev+12, %edx
addl$1, %eax
adcl$0, %edx
movl%eax, __gcov0._ZN4TestC2Ev+8
movl%edx, __gcov0._ZN4TestC2Ev+12
subl$8, %esp
pushl   $_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_
pushl   %ecx
call_ZNSolsEPFRSoS_E
addl$16, %esp
movl__gcov0._ZN4TestC2Ev+16, %eax
movl__gcov0._ZN4TestC2Ev+20, %edx
addl$1, %eax
adcl$0, %edx
movl%eax, __gcov0._ZN4TestC2Ev+16
movl%edx, __gcov0._ZN4TestC2Ev+20
nop
leave
.cfi_restore 5
.cfi_def_cfa 4, 4
ret
.cfi_endproc

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 09/30/2016 02:31 PM, Rainer Orth wrote:

this would be i386-pc-solaris2.12.  I'm not sure if the constructor
priority detection works in a cross scenario.


Hi.

By the way, I tried to test the cross-compiler:
$ ../configure --disable-bootstrap --enable-languages=c,c++,fortran 
--enable-valgrind-annotations --prefix=/home/marxin/bin/gcc2 --disable-multilib 
--disable-libsanitizer --target=i386-pc-solaris2.12

and I get for:
cat /tmp/priority.c
void __attribute__ ((constructor(150))) foo()
{
}

void __attribute__ ((constructor(151))) bar()
{
}

int main()
{
  return 0;
}

$ ./xgcc -B. /tmp/priority.c -fprofile-generate -S
/tmp/priority.c:2:1: error: constructor priorities are not supported
 {
 ^
/tmp/priority.c:6:1: error: constructor priorities are not supported
 {
 ^

I guess even cross compiler should detect whether the target supports ctor/dtor 
priorities.
May I ask you for assembly file of a native compiler with the suggested patch?

Thanks,
Martin

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 10/03/2016 03:03 PM, Rainer Orth wrote:
> Hi Martin,
> 
>> On 09/30/2016 02:31 PM, Rainer Orth wrote:
>>> this would be i386-pc-solaris2.12.  I'm not sure if the constructor
>>> priority detection works in a cross scenario.
>>>
>>> I'm attaching the resulting assembly (although for Solaris as, the gas
>>> build is still running).
>>
>> Hi. Sorry, I have a stupid mistake in dtor priority
>> (I used 65534 instead of desired 99). Please try to test it on Solaris 12
>> with the attached patch. I'll send the patch to ML soon.
> 
> unfortunately, the patch makes no difference on Solaris 12.  The test
> even FAILs when using gas/gld, which is a different/independent
> implementation of constructor priority.

Ok, can you please send me x.S file for Solaris 12?

> 
>> Can you please test whether it makes any change on a solaris target w/o
>> prioritized ctors/dtors?
> 
> It doesn't: the test PASSes on Solaris 10 and 11 with and without your
> patch.

I see, that would require the former approach using atexit, which would be
chosen depending on whether target supports prioritized dtors or not.

Martin

> 
>   Rainer
> 

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

> On 09/30/2016 02:31 PM, Rainer Orth wrote:
>> this would be i386-pc-solaris2.12.  I'm not sure if the constructor
>> priority detection works in a cross scenario.
>> 
>> I'm attaching the resulting assembly (although for Solaris as, the gas
>> build is still running).
>
> Hi. Sorry, I have a stupid mistake in dtor priority
> (I used 65534 instead of desired 99). Please try to test it on Solaris 12
> with the attached patch. I'll send the patch to ML soon.

unfortunately, the patch makes no difference on Solaris 12.  The test
even FAILs when using gas/gld, which is a different/independent
implementation of constructor priority.

> Can you please test whether it makes any change on a solaris target w/o
> prioritized ctors/dtors?

It doesn't: the test PASSes on Solaris 10 and 11 with and without your
patch.

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

> On 09/30/2016 02:31 PM, Rainer Orth wrote:
>> this would be i386-pc-solaris2.12.  I'm not sure if the constructor
>> priority detection works in a cross scenario.
>> 
>> I'm attaching the resulting assembly (although for Solaris as, the gas
>> build is still running).
>
> Hi. Sorry, I have a stupid mistake in dtor priority
> (I used 65534 instead of desired 99). Please try to test it on Solaris 12
> with the attached patch. I'll send the patch to ML soon.
>
> Can you please test whether it makes any change on a solaris target w/o
> prioritized ctors/dtors?

sure: I've added your patch to my source tree for the running bootstraps
and will have builds on Solaris 12 (with constructor priority) and S10/11
(without) available in a few hours.

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 09/30/2016 02:31 PM, Rainer Orth wrote:
> this would be i386-pc-solaris2.12.  I'm not sure if the constructor
> priority detection works in a cross scenario.
> 
> I'm attaching the resulting assembly (although for Solaris as, the gas
> build is still running).

Hi. Sorry, I have a stupid mistake in dtor priority
(I used 65534 instead of desired 99). Please try to test it on Solaris 12
with the attached patch. I'll send the patch to ML soon.

Can you please test whether it makes any change on a solaris target w/o
prioritized ctors/dtors?

Thanks,
Martin
diff --git a/gcc/coverage.c b/gcc/coverage.c
index 0b8c0b3..a759831 100644
--- a/gcc/coverage.c
+++ b/gcc/coverage.c
@@ -1078,7 +1078,7 @@ build_gcov_exit_decl (void)
   append_to_statement_list (stmt, );
 
   /* Generate a destructor to run it (with priority 99).  */
-  cgraph_build_static_cdtor ('D', dtor, DEFAULT_INIT_PRIORITY - 1);
+  cgraph_build_static_cdtor ('D', dtor, MAX_RESERVED_INIT_PRIORITY - 1);
 }
 
 /* Create the gcov_info types and object.  Generate the constructor

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

>> understood.  However, Solaris 12 *does* have support for constructor
>> priorities and the testcase still fails, so there's more going on here.
>
> I see, however I don't have access to such a machine. I would appreciate
> if you help me to debug what's going on. Can you please send me --target=x,
> so that I can at least check created assembly?

this would be i386-pc-solaris2.12.  I'm not sure if the constructor
priority detection works in a cross scenario.

I'm attaching the resulting assembly (although for Solaris as, the gas
build is still running).

Here's the gcov -b pr16855.C output

File '/vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/gcov/pr16855.C'
Lines executed:73.91% of 23
Branches executed:100.00% of 4
Taken at least once:50.00% of 4
Calls executed:71.43% of 14
Creating 'pr16855.C.gcov'

File 
'/var/gcc/regression/trunk/12-gcc/build/i386-pc-solaris2.12/amd64/libstdc++-v3/include/iostream'
Lines executed:100.00% of 1
No branches
Calls executed:100.00% of 2
Creating 'iostream.gcov'

File 
'/var/gcc/regression/trunk/12-gcc/build/i386-pc-solaris2.12/amd64/libstdc++-v3/include/i386-pc-solaris2.12/bits/gthr-default.h'
Lines executed:0.00% of 3
No branches
No calls
Creating 'gthr-default.h.gcov'


compared to what I get on Linux/x86_64:

File '/vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/gcov/pr16855.C'
Lines executed:86.96% of 23
Branches executed:100.00% of 4
Taken at least once:50.00% of 4
Calls executed:85.71% of 14
Creating 'pr16855.C.gcov'

File 
'/var/gcc/regression/trunk/4.7.4-gcc-gas-gld/build/x86_64-pc-linux-gnu/libstdc++-v3/include/iostream'
Lines executed:100.00% of 1
No branches
Calls executed:100.00% of 2
Creating 'iostream.gcov'

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University


.file   "pr16855.C"
.globl  a
.bss
.align 4
.type   a, @object
.size   a, 4
a:
.zero   4
.text
.globl  _Z3foov
.type   _Z3foov, @function
_Z3foov:
.LFB5:
pushq   %rbp
.LCFI0:
movq%rsp, %rbp
.LCFI1:
movq__gcov0._Z3foov(%rip), %rax
addq$1, %rax
movq%rax, __gcov0._Z3foov(%rip)
movl$123, a(%rip)
nop
popq%rbp
.LCFI2:
ret
.LFE5:
.size   _Z3foov, .-_Z3foov
.data
.align 4
.type   _ZL16__gthread_active, @object
.size   _ZL16__gthread_active, 4
_ZL16__gthread_active:
.long   -1
.text
.type   _ZL17__gthread_triggerv, @function
_ZL17__gthread_triggerv:
.LFB286:
pushq   %rbp
.LCFI3:
movq%rsp, %rbp
.LCFI4:
movq__gcov0._ZL17__gthread_triggerv(%rip), %rax
addq$1, %rax
movq%rax, __gcov0._ZL17__gthread_triggerv(%rip)
movl$1, _ZL16__gthread_active(%rip)
nop
popq%rbp
.LCFI5:
ret
.LFE286:
.size   _ZL17__gthread_triggerv, .-_ZL17__gthread_triggerv
.local  _ZStL8__ioinit
.comm   _ZStL8__ioinit,1,1
.section.rodata
.LC0:
.string "In Test ctor"
.section.text._ZN4TestC2Ev%_ZN4TestC5Ev,"ax",@progbits
.group  _ZN4TestC5Ev,.text._ZN4TestC2Ev%_ZN4TestC5Ev,#comdat
.align 2
.weak   _ZN4TestC2Ev
.type   _ZN4TestC2Ev, @function
_ZN4TestC2Ev:
.LFB1036:
pushq   %rbp
.LCFI6:
movq%rsp, %rbp
.LCFI7:
subq$16, %rsp
movq%rdi, -8(%rbp)
movq__gcov0._ZN4TestC2Ev(%rip), %rax
addq$1, %rax
movq%rax, __gcov0._ZN4TestC2Ev(%rip)
movl$.LC0, %esi
movl$_ZSt4cout, %edi
call_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_PKc
movq%rax, %rdx
movq__gcov0._ZN4TestC2Ev+8(%rip), %rax
addq$1, %rax
movq%rax, __gcov0._ZN4TestC2Ev+8(%rip)
movl$_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_, 
%esi
movq%rdx, %rdi
call_ZNSolsEPFRSoS_E
movq__gcov0._ZN4TestC2Ev+16(%rip), %rax
addq$1, %rax
movq%rax, __gcov0._ZN4TestC2Ev+16(%rip)
nop
leave
.LCFI8:
ret
.LFE1036:
.size   _ZN4TestC2Ev, .-_ZN4TestC2Ev
.weak   _ZN4TestC1Ev
.set_ZN4TestC1Ev,_ZN4TestC2Ev
.section.rodata
.LC1:
.string "In Test dtor"
.section.text._ZN4TestD2Ev%_ZN4TestD5Ev,"ax",@progbits
.group  _ZN4TestD5Ev,.text._ZN4TestD2Ev%_ZN4TestD5Ev,#comdat
.align 2
.weak   _ZN4TestD2Ev
.type   _ZN4TestD2Ev, @function
_ZN4TestD2Ev:
.LFB1039:
pushq   %rbp
.LCFI9:
movq%rsp, %rbp
.LCFI10:
subq$16, %rsp
movq%rdi, -8(%rbp)
movq__gcov0._ZN4TestD2Ev(%rip), %rax
addq$1, %rax
movq%rax, __gcov0._ZN4TestD2Ev(%rip)
movl$.LC1, %esi

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Nathan Sidwell]

On 09/30/16 05:22, Rainer Orth wrote:


While this would take care of the testsuite failures, this creates a
terrible user experience outside of the testsuite: if we know that
-fprofile-arcs -ftest-coverage cannot work on targets without
constructor priority support, the compiler should error out with an
appropriate message instead of just creating confusing non-working
executables.


It should either
1) emit a non-prioritized static ctor
2) or use the older atexit mechanism.

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 09/30/2016 11:22 AM, Rainer Orth wrote:
> Hi Martin,
> 
>>> the testcase FAILs on Solaris 12 (both SPARC and x86):
>>>
>>> +FAIL: g++.dg/gcov/pr16855.C -std=gnu++11 gcov: 1 failures in line
>>> counts, 0 i
>>> n branch percentages, 0 in return percentages, 0 in intermediate format
>>> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++11  line 21: is #:should be 1
>>> +FAIL: g++.dg/gcov/pr16855.C -std=gnu++14 gcov: 1 failures in line
>>> counts, 0 i
>>> n branch percentages, 0 in return percentages, 0 in intermediate format
>>> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++14  line 21: is #:should be 1
>>> +FAIL: g++.dg/gcov/pr16855.C -std=gnu++98 gcov: 1 failures in line
>>> counts, 0 i
>>> n branch percentages, 0 in return percentages, 0 in intermediate format
>>> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++98  line 21: is #:should be 1
>>>
>>> I haven't looked closer yet, but notice that you require constructor
>>> priority support which isn't available everywhere (it is on Solaris 12,
>>> but not before).
>>>
>>> Rainer
>>>
>>
>> Hello.
>>
>> Sorry for the test-breakage. The issue is really connected to fact that
>> current trunk relies
>> on support of dtor priority. The former implementation called the function
>> __gcov_exit via atexit.
>> If I understand correctly, fully support of static ctors/dtors, C++
>> ctors/dtors, with combination
>> of atexit cannot be done on a target w/o ctor/dtor priorities.
> 
> understood.  However, Solaris 12 *does* have support for constructor
> priorities and the testcase still fails, so there's more going on here.

I see, however I don't have access to such a machine. I would appreciate
if you help me to debug what's going on. Can you please send me --target=x,
so that I can at least check created assembly?

> 
>> Ideally we should have a macro for each target telling whether it supports
>> priorities or not.
>> However, we probably don't have? I would suggest to make the test
>> conditional just for main
>> targets which support priorities?
>>
>> Thoughts?
> 
> While this would take care of the testsuite failures, this creates a
> terrible user experience outside of the testsuite: if we know that
> -fprofile-arcs -ftest-coverage cannot work on targets without
> constructor priority support, the compiler should error out with an
> appropriate message instead of just creating confusing non-working
> executables.

More precisely, it does not work reliably on constructor and destructors as
we depend on an order how are ctor/dtors executed. We had the same behavior even
before my patch, but documenting that definitely worth for doing. I'll do a 
patch.

Martin

> 
>   Rainer
> 

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

>> the testcase FAILs on Solaris 12 (both SPARC and x86):
>> 
>> +FAIL: g++.dg/gcov/pr16855.C -std=gnu++11 gcov: 1 failures in line
>> counts, 0 i
>> n branch percentages, 0 in return percentages, 0 in intermediate format
>> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++11  line 21: is #:should be 1
>> +FAIL: g++.dg/gcov/pr16855.C -std=gnu++14 gcov: 1 failures in line
>> counts, 0 i
>> n branch percentages, 0 in return percentages, 0 in intermediate format
>> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++14  line 21: is #:should be 1
>> +FAIL: g++.dg/gcov/pr16855.C -std=gnu++98 gcov: 1 failures in line
>> counts, 0 i
>> n branch percentages, 0 in return percentages, 0 in intermediate format
>> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++98  line 21: is #:should be 1
>> 
>> I haven't looked closer yet, but notice that you require constructor
>> priority support which isn't available everywhere (it is on Solaris 12,
>> but not before).
>> 
>>  Rainer
>> 
>
> Hello.
>
> Sorry for the test-breakage. The issue is really connected to fact that
> current trunk relies
> on support of dtor priority. The former implementation called the function
> __gcov_exit via atexit.
> If I understand correctly, fully support of static ctors/dtors, C++
> ctors/dtors, with combination
> of atexit cannot be done on a target w/o ctor/dtor priorities.

understood.  However, Solaris 12 *does* have support for constructor
priorities and the testcase still fails, so there's more going on here.

> Ideally we should have a macro for each target telling whether it supports
> priorities or not.
> However, we probably don't have? I would suggest to make the test
> conditional just for main
> targets which support priorities?
>
> Thoughts?

While this would take care of the testsuite failures, this creates a
terrible user experience outside of the testsuite: if we know that
-fprofile-arcs -ftest-coverage cannot work on targets without
constructor priority support, the compiler should error out with an
appropriate message instead of just creating confusing non-working
executables.

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 09/29/2016 03:00 PM, Nathan Sidwell wrote:
> On 09/29/16 08:54, Nathan Sidwell wrote:
>> On 09/29/16 08:49, Martin Liška wrote:
>>> Ideally we should have a macro for each target telling whether it supports
>>> priorities or not.
>>> However, we probably don't have? I would suggest to make the test 
>>> conditional
>>> just for main
>>> targets which support priorities?
>>
>> or a dg_effective_target test.  Probably overkill if there's exactly one 
>> target
>> impacted.
> 
> already there : effective_target_init_priority

Nice that we have it. Looks it's going to be very first usage of the effective 
target.
I'm suggesting patch which also changes debugging messages to make it more 
readable.

Ready for trunk?
Thanks,
Martin
>From cd66f6f9b9c68267698720dcf350b140d86f4201 Mon Sep 17 00:00:00 2001
From: marxin 
Date: Thu, 29 Sep 2016 15:39:08 +0200
Subject: [PATCH] Use effective target for pr16855

gcc/testsuite/ChangeLog:

2016-09-29  Martin Liska  

	* g++.dg/gcov/pr16855.C: Add init_priority as an effective
	target.
---
 gcc/testsuite/g++.dg/gcov/pr16855.C | 15 ---
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/gcc/testsuite/g++.dg/gcov/pr16855.C b/gcc/testsuite/g++.dg/gcov/pr16855.C
index 91801d4..8167176 100644
--- a/gcc/testsuite/g++.dg/gcov/pr16855.C
+++ b/gcc/testsuite/g++.dg/gcov/pr16855.C
@@ -1,13 +1,15 @@
 /* { dg-options "-fprofile-arcs -ftest-coverage" } */
 /* { dg-do run { target native } } */
+/* { dg-require-effective-target init_priority  } */
 
 #include 
+#include 
 
 int a;
 
 void foo()
 {
-  a = 123;		  /* count(1) */
+  fprintf (stderr, "atexit handler foo()\n"); 		  /* count(1) */
 }
 
 #include 
@@ -15,10 +17,10 @@ using namespace std;
 class Test {
 public:
 	Test(void){
-	cout<< "In Test ctor" << endl;			  /* count(1) */
+	fprintf (stderr, "In Test ctor\n");		  /* count(1) */
 	}
 	~Test(void){
-	cout<< "In Test dtor" << endl;			  /* count(1) */
+	fprintf (stderr, "In Test dtor\n");		  /* count(1) */
 	}
 }T1;
 
@@ -27,8 +29,7 @@ void uncalled(void){
 }
 int main(void){
 atexit ();
-// Test T2;
-cout<< "In main" << endl;  /* count(1) */
+fprintf (stderr, "In main\n");  /* count(1) */
 return 0;
 }
 
@@ -36,12 +37,12 @@ return 0;
 
 __attribute__((constructor))
 static void construct_navigationBarImages() {
-  fprintf (stderr,  "((construct_navigationBarImages))"); /* count(1) */
+  fprintf (stderr,  "((construct_navigationBarImages))\n"); /* count(1) */
 }
 
 __attribute__((destructor))
 static void destroy_navigationBarImages() {
-  fprintf (stderr,  "((destroy_navigationBarImages))");	  /* count(1) */
+  fprintf (stderr,  "((destroy_navigationBarImages))\n");  /* count(1) */
 }
 
 /* { dg-final { run-gcov branches { -b pr16855.C } } } */
-- 
2.9.2

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Nathan Sidwell]

On 09/29/16 08:54, Nathan Sidwell wrote:

On 09/29/16 08:49, Martin Liška wrote:

Ideally we should have a macro for each target telling whether it supports
priorities or not.
However, we probably don't have? I would suggest to make the test conditional
just for main
targets which support priorities?


or a dg_effective_target test.  Probably overkill if there's exactly one target
impacted.


already there : effective_target_init_priority

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Nathan Sidwell]

On 09/29/16 08:49, Martin Liška wrote:

Ideally we should have a macro for each target telling whether it supports 
priorities or not.
However, we probably don't have? I would suggest to make the test conditional 
just for main
targets which support priorities?


or a dg_effective_target test.  Probably overkill if there's exactly one target 
impacted.


nathan

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 09/29/2016 11:00 AM, Rainer Orth wrote:
> Hi Martin,
> 
>> On 08/12/2016 04:08 PM, Martin Liška wrote:
>>> On 08/10/2016 02:53 PM, Nathan Sidwell wrote:
 On 08/10/16 06:43, Martin Liška wrote:
> Hello.
>
> There are multiple PRs (mentioned in ChangeLog) which suffer from
> missing capability of gcov
> to save counters for functions with constructor/destructor
> attributes. I done that by simply
> replacing atexit handler (gcov_exit) with a new static destructor
> (__gcov_exit), which has
> priority 99 (I did the same for __gcov_init). However, I'm not sure
> whether it's possible
> that a ctor defined in a source file can be potentially executed before
> __gcov_init (w/ the default
> priority)?
>
> Patch survives:
> make check -k -j10 RUNTESTFLAGS="gcov.exp"
> make check -k -j10 RUNTESTFLAGS="tree-prof.exp"
>
> I've just also verified that a DSO gcov dump works as before. I'm
> attaching a test-case which
> tests both static ctors/dtors, as well as C++ ctors/dtors.

 Does a coverage bootstrap (--enable-coverage) still succeed?
>>>
>>> Well, looks results are more unstable than I thought.
>>> Even running 'make -j1' in objdir/x86_64-pc-linux-gnu/libgcc repeatedly
>>> generates different results.
>>> I'll dig in after weekend.
>>>
>>> Martin
> [...]
 I think this is a good idea, but we should document the changed
 behavior. (I don't think the current behaviour's documented).
>>
>> I'm adding a new hunk that documents the behavior.
>>
>> Is the patch ready to be installed?
> 
> the testcase FAILs on Solaris 12 (both SPARC and x86):
> 
> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++11  gcov: 1 failures in line counts, 
> 0 i
> n branch percentages, 0 in return percentages, 0 in intermediate format
> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++11  line 21: is #:should be 1
> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++14  gcov: 1 failures in line counts, 
> 0 i
> n branch percentages, 0 in return percentages, 0 in intermediate format
> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++14  line 21: is #:should be 1
> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++98  gcov: 1 failures in line counts, 
> 0 i
> n branch percentages, 0 in return percentages, 0 in intermediate format
> +FAIL: g++.dg/gcov/pr16855.C  -std=gnu++98  line 21: is #:should be 1
> 
> I haven't looked closer yet, but notice that you require constructor
> priority support which isn't available everywhere (it is on Solaris 12,
> but not before).
> 
>   Rainer
> 

Hello.

Sorry for the test-breakage. The issue is really connected to fact that current 
trunk relies
on support of dtor priority. The former implementation called the function 
__gcov_exit via atexit.
If I understand correctly, fully support of static ctors/dtors, C++ 
ctors/dtors, with combination
of atexit cannot be done on a target w/o ctor/dtor priorities.

Ideally we should have a macro for each target telling whether it supports 
priorities or not.
However, we probably don't have? I would suggest to make the test conditional 
just for main
targets which support priorities?

Thoughts?

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Rainer Orth]

Hi Martin,

> On 08/12/2016 04:08 PM, Martin Liška wrote:
>> On 08/10/2016 02:53 PM, Nathan Sidwell wrote:
>>> On 08/10/16 06:43, Martin Liška wrote:
 Hello.

 There are multiple PRs (mentioned in ChangeLog) which suffer from
 missing capability of gcov
 to save counters for functions with constructor/destructor
 attributes. I done that by simply
 replacing atexit handler (gcov_exit) with a new static destructor
 (__gcov_exit), which has
 priority 99 (I did the same for __gcov_init). However, I'm not sure
 whether it's possible
 that a ctor defined in a source file can be potentially executed before
 __gcov_init (w/ the default
 priority)?

 Patch survives:
 make check -k -j10 RUNTESTFLAGS="gcov.exp"
 make check -k -j10 RUNTESTFLAGS="tree-prof.exp"

 I've just also verified that a DSO gcov dump works as before. I'm
 attaching a test-case which
 tests both static ctors/dtors, as well as C++ ctors/dtors.
>>>
>>> Does a coverage bootstrap (--enable-coverage) still succeed?
>> 
>> Well, looks results are more unstable than I thought.
>> Even running 'make -j1' in objdir/x86_64-pc-linux-gnu/libgcc repeatedly
>> generates different results.
>> I'll dig in after weekend.
>> 
>> Martin
[...]
>>> I think this is a good idea, but we should document the changed
>>> behavior. (I don't think the current behaviour's documented).
>
> I'm adding a new hunk that documents the behavior.
>
> Is the patch ready to be installed?

the testcase FAILs on Solaris 12 (both SPARC and x86):

+FAIL: g++.dg/gcov/pr16855.C  -std=gnu++11  gcov: 1 failures in line counts, 0 i
n branch percentages, 0 in return percentages, 0 in intermediate format
+FAIL: g++.dg/gcov/pr16855.C  -std=gnu++11  line 21: is #:should be 1
+FAIL: g++.dg/gcov/pr16855.C  -std=gnu++14  gcov: 1 failures in line counts, 0 i
n branch percentages, 0 in return percentages, 0 in intermediate format
+FAIL: g++.dg/gcov/pr16855.C  -std=gnu++14  line 21: is #:should be 1
+FAIL: g++.dg/gcov/pr16855.C  -std=gnu++98  gcov: 1 failures in line counts, 0 i
n branch percentages, 0 in return percentages, 0 in intermediate format
+FAIL: g++.dg/gcov/pr16855.C  -std=gnu++98  line 21: is #:should be 1

I haven't looked closer yet, but notice that you require constructor
priority support which isn't available everywhere (it is on Solaris 12,
but not before).

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Nathan Sidwell]

On 09/27/16 07:07, Martin Liška wrote:

On 09/27/2016 12:55 PM, Nathan Sidwell wrote:


"Instrumented applications use a static destructor with priority 99 to invoke 
the __gcov_dump function. Thus __gcov_dump is executed after all
user defined static destructors, as well as handlers registered with atexit."

?


Hello.

I like your wording, I installed the patch as r240529.


thanks!

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 09/27/2016 12:55 PM, Nathan Sidwell wrote:
> 
> "Instrumented applications use a static destructor with priority 99 to invoke 
> the __gcov_dump function. Thus __gcov_dump is executed after all
> user defined static destructors, as well as handlers registered with atexit."
> 
> ?

Hello.

I like your wording, I installed the patch as r240529.

Martin

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Nathan Sidwell]

On 09/26/16 11:22, Martin Liška wrote:


Hi.

So the I found reason of inconsistencies, running multiple times -fselftest is 
enough to
find that memory allocation related functions can be executed different times.
Small example:


thanks for checking.

@@ -598,6 +598,10 @@ facilities to restrict profile collection to the program 
region of

 interest. Calling @code{__gcov_reset(void)} will clear all profile counters
 to zero, and calling @code{__gcov_dump(void)} will cause the profile 
information
 collected at that point to be dumped to @file{.gcda} output files.
+By default, every instrumented application calls __gcov_dump function
+via a static destructor with priority equal to 99.  That would guarantee
+that all user defined destructors, as well as function handlers registered
+by atexit, would be executed before gcov dump function is executed.

'by default'  This suggests there's a non-default behaviour, but I can't see it 
nor how to enable it.  Perhaps:


"Instrumented applications use a static destructor with priority 99 to invoke 
the __gcov_dump function. Thus __gcov_dump is executed after all

user defined static destructors, as well as handlers registered with atexit."

?

nathan

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 08/12/2016 04:08 PM, Martin Liška wrote:
> On 08/10/2016 02:53 PM, Nathan Sidwell wrote:
>> On 08/10/16 06:43, Martin Liška wrote:
>>> Hello.
>>>
>>> There are multiple PRs (mentioned in ChangeLog) which suffer from missing 
>>> capability of gcov
>>> to save counters for functions with constructor/destructor attributes. I 
>>> done that by simply
>>> replacing atexit handler (gcov_exit) with a new static destructor 
>>> (__gcov_exit), which has
>>> priority 99 (I did the same for __gcov_init). However, I'm not sure whether 
>>> it's possible
>>> that a ctor defined in a source file can be potentially executed before 
>>> __gcov_init (w/ the default
>>> priority)?
>>>
>>> Patch survives:
>>> make check -k -j10 RUNTESTFLAGS="gcov.exp"
>>> make check -k -j10 RUNTESTFLAGS="tree-prof.exp"
>>>
>>> I've just also verified that a DSO gcov dump works as before. I'm attaching 
>>> a test-case which
>>> tests both static ctors/dtors, as well as C++ ctors/dtors.
>>
>> Does a coverage bootstrap (--enable-coverage) still succeed?
> 
> Well, looks results are more unstable than I thought.
> Even running 'make -j1' in objdir/x86_64-pc-linux-gnu/libgcc repeatedly 
> generates different results.
> I'll dig in after weekend.
> 
> Martin

Hi.

So the I found reason of inconsistencies, running multiple times -fselftest is 
enough to
find that memory allocation related functions can be executed different times.
Small example:

--- /tmp/r1/ggc-page.c.gcov 2016-09-26 16:54:53.060921496 +0200
+++ /tmp/r2/ggc-page.c.gcov 2016-09-26 16:55:52.470058958 +0200
@@ -636,8 +636,8 @@
 -:  631:#else
307718:  632:  page_table table = G.lookup;
307718:  633:  uintptr_t high_bits = (uintptr_t) p & ~ (uintptr_t) 
0x;
-   307718:  634:  while (table->high_bits != high_bits)
-#:  635:table = table->next;
+   307794:  634:  while (table->high_bits != high_bits)
+   38:  635:table = table->next;
307718:  636:  base = >table[0];
 -:  637:#endif
 -:  638:
@@ -661,15 +661,15 @@
 -:  656:#else
 -:  657:  page_table table;
  2134:  658:  uintptr_t high_bits = (uintptr_t) p & ~ (uintptr_t) 
0x;
- 2134:  659:  for (table = G.lookup; table; table = table->next)
- 2133:  660:if (table->high_bits == high_bits)
- 2133:  661:  goto found;
+ 2322:  659:  for (table = G.lookup; table; table = table->next)
+ 2320:  660:if (table->high_bits == high_bits)
+ 2132:  661:  goto found;
 -:  662:
 -:  663:  /* Not found -- allocate a new table.  */
-1:  664:  table = XCNEW (struct page_table_chain);
-1:  665:  table->next = G.lookup;
-1:  666:  table->high_bits = high_bits;
-1:  667:  G.lookup = table;
+2:  664:  table = XCNEW (struct page_table_chain);
+2:  665:  table->next = G.lookup;
+2:  666:  table->high_bits = high_bits;
+2:  667:  G.lookup = table;
  2134:  668:found:
  2134:  669:  base = >table[0];
 -:  670:#endif
--- /tmp/r1/ggc-page.c.gcov 2016-09-26 16:54:53.060921496 +0200
+++ /tmp/r2/ggc-page.c.gcov 2016-09-26 16:58:22.440931009 +0200
@@ -679,7 +679,7 @@
  2134:  674:  L2 = LOOKUP_L2 (p);
 -:  675:
  2134:  676:  if (base[L1] == NULL)
-3:  677:base[L1] = XCNEWVEC (page_entry *, PAGE_L2_SIZE);
+2:  677:base[L1] = XCNEWVEC (page_entry *, PAGE_L2_SIZE);
 -:  678:
  2134:  679:  base[L1][L2] = entry;
  2134:  680:}

It's reasonable to me that it can change. However, the patch I would like to 
install
does not cause any new differences.

Martin

> 
>>
>> I think this is a good idea, but we should document the changed behavior. (I 
>> don't think the current behaviour's documented).

I'm adding a new hunk that documents the behavior.

Is the patch ready to be installed?
Thanks,
Martin

>>
>>
>> nathan
> 

>From 686e65a923288c2c5055a9edb61e6f0648d6a2a3 Mon Sep 17 00:00:00 2001
From: marxin 
Date: Wed, 10 Aug 2016 12:18:45 +0200
Subject: [PATCH] gcov: dump in a static dtor instead of in an atexit handler

gcc/testsuite/ChangeLog:

2016-08-10  Martin Liska  

	PR gcov-profile/7970
	PR gcov-profile/16855
	PR gcov-profile/44779
	* g++.dg/gcov/pr16855.C: New test.

gcc/ChangeLog:

2016-08-10  Martin Liska  

	PR gcov-profile/7970
	PR gcov-profile/16855
	PR gcov-profile/44779
	* coverage.c (build_gcov_exit_decl): New function.
	(coverage_obj_init): Call the function and generate __gcov_exit
	destructor.
	* doc/gcov.texi: Document when __gcov_exit function is called.

libgcc/ChangeLog:

2016-08-10  Martin Liska  

	PR gcov-profile/7970
	PR gcov-profile/16855
	PR gcov-profile/44779
	* libgcov-driver.c (__gcov_init): Do not register a atexit
	handler.
	(__gcov_exit): Rename from gcov_exit.
	* libgcov.h (__gcov_exit): Declare.
---
 gcc/coverage.c  | 27 +++--
 

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

On 08/10/2016 02:53 PM, Nathan Sidwell wrote:
> On 08/10/16 06:43, Martin Liška wrote:
>> Hello.
>>
>> There are multiple PRs (mentioned in ChangeLog) which suffer from missing 
>> capability of gcov
>> to save counters for functions with constructor/destructor attributes. I 
>> done that by simply
>> replacing atexit handler (gcov_exit) with a new static destructor 
>> (__gcov_exit), which has
>> priority 99 (I did the same for __gcov_init). However, I'm not sure whether 
>> it's possible
>> that a ctor defined in a source file can be potentially executed before 
>> __gcov_init (w/ the default
>> priority)?
>>
>> Patch survives:
>> make check -k -j10 RUNTESTFLAGS="gcov.exp"
>> make check -k -j10 RUNTESTFLAGS="tree-prof.exp"
>>
>> I've just also verified that a DSO gcov dump works as before. I'm attaching 
>> a test-case which
>> tests both static ctors/dtors, as well as C++ ctors/dtors.
> 
> Does a coverage bootstrap (--enable-coverage) still succeed?

Well, looks results are more unstable than I thought.
Even running 'make -j1' in objdir/x86_64-pc-linux-gnu/libgcc repeatedly 
generates different results.
I'll dig in after weekend.

Martin

> 
> I think this is a good idea, but we should document the changed behavior. (I 
> don't think the current behaviour's documented).
> 
> 
> nathan

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Nathan Sidwell]

On 08/10/16 08:53, Nathan Sidwell wrote:


I think this is a good idea, but we should document the changed behavior. (I
don't think the current behaviour's documented).


oh, gcov_exit is a user callable routine.  You'll have to keep it available.

Re: [PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Nathan Sidwell]

On 08/10/16 06:43, Martin Liška wrote:

Hello.

There are multiple PRs (mentioned in ChangeLog) which suffer from missing 
capability of gcov
to save counters for functions with constructor/destructor attributes. I done 
that by simply
replacing atexit handler (gcov_exit) with a new static destructor 
(__gcov_exit), which has
priority 99 (I did the same for __gcov_init). However, I'm not sure whether 
it's possible
that a ctor defined in a source file can be potentially executed before 
__gcov_init (w/ the default
priority)?

Patch survives:
make check -k -j10 RUNTESTFLAGS="gcov.exp"
make check -k -j10 RUNTESTFLAGS="tree-prof.exp"

I've just also verified that a DSO gcov dump works as before. I'm attaching a 
test-case which
tests both static ctors/dtors, as well as C++ ctors/dtors.


Does a coverage bootstrap (--enable-coverage) still succeed?

I think this is a good idea, but we should document the changed behavior. (I 
don't think the current behaviour's documented).



nathan

[PATCH, RFC] gcov: dump in a static dtor instead of in an atexit handler

[Martin Liška]

Hello.

There are multiple PRs (mentioned in ChangeLog) which suffer from missing 
capability of gcov
to save counters for functions with constructor/destructor attributes. I done 
that by simply
replacing atexit handler (gcov_exit) with a new static destructor 
(__gcov_exit), which has
priority 99 (I did the same for __gcov_init). However, I'm not sure whether 
it's possible
that a ctor defined in a source file can be potentially executed before 
__gcov_init (w/ the default
priority)?

Patch survives:
make check -k -j10 RUNTESTFLAGS="gcov.exp"
make check -k -j10 RUNTESTFLAGS="tree-prof.exp"

I've just also verified that a DSO gcov dump works as before. I'm attaching a 
test-case which 
tests both static ctors/dtors, as well as C++ ctors/dtors.

Thoughts?
Martin
>From 0e7f660b77628533679e6302a3f4b444166fc365 Mon Sep 17 00:00:00 2001
From: marxin 
Date: Wed, 10 Aug 2016 12:18:45 +0200
Subject: [PATCH] gcov: dump in a static dtor instead of in an atexit handler

gcc/testsuite/ChangeLog:

2016-08-10  Martin Liska  

	PR gcov-profile/7970
	PR gcov-profile/16855
	PR gcov-profile/44779
	* g++.dg/gcov/pr16855.C: New test.

gcc/ChangeLog:

2016-08-10  Martin Liska  

	PR gcov-profile/7970
	PR gcov-profile/16855
	PR gcov-profile/44779
	* coverage.c (build_gcov_exit_decl): New function.
	(coverage_obj_init): Call the function and generate __gcov_exit
	destructor.

libgcc/ChangeLog:

2016-08-10  Martin Liska  

	PR gcov-profile/7970
	PR gcov-profile/16855
	PR gcov-profile/44779
	* libgcov-driver.c (__gcov_init): Do not register a atexit
	handler.
	(__gcov_exit): Rename from gcov_exit.
	* libgcov.h (__gcov_exit): Declare.
---
 gcc/coverage.c  | 27 +--
 gcc/testsuite/g++.dg/gcov/pr16855.C | 37 +
 libgcc/libgcov-driver.c |  5 ++---
 libgcc/libgcov.h|  3 +++
 4 files changed, 67 insertions(+), 5 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/gcov/pr16855.C

diff --git a/gcc/coverage.c b/gcc/coverage.c
index d4d371e..da7f915 100644
--- a/gcc/coverage.c
+++ b/gcc/coverage.c
@@ -1054,8 +1054,30 @@ build_init_ctor (tree gcov_info_type)
   stmt = build_call_expr (init_fn, 1, stmt);
   append_to_statement_list (stmt, );
 
-  /* Generate a constructor to run it.  */
-  cgraph_build_static_cdtor ('I', ctor, DEFAULT_INIT_PRIORITY);
+  /* Generate a constructor to run it (with priority 99).  */
+  cgraph_build_static_cdtor ('I', ctor, DEFAULT_INIT_PRIORITY - 1);
+}
+
+/* Generate the destructor function to call __gcov_exit.  */
+
+static void
+build_gcov_exit_decl (void)
+{
+  tree init_fn = build_function_type_list (void_type_node, void_type_node,
+	   NULL);
+  init_fn = build_decl (BUILTINS_LOCATION, FUNCTION_DECL,
+			get_identifier ("__gcov_exit"), init_fn);
+  TREE_PUBLIC (init_fn) = 1;
+  DECL_EXTERNAL (init_fn) = 1;
+  DECL_ASSEMBLER_NAME (init_fn);
+
+  /* Generate a call to __gcov_exit ().  */
+  tree dtor = NULL;
+  tree stmt = build_call_expr (init_fn, 0);
+  append_to_statement_list (stmt, );
+
+  /* Generate a destructor to run it (with priority 99).  */
+  cgraph_build_static_cdtor ('D', dtor, DEFAULT_INIT_PRIORITY - 1);
 }
 
 /* Create the gcov_info types and object.  Generate the constructor
@@ -1113,6 +1135,7 @@ coverage_obj_init (void)
   DECL_NAME (gcov_info_var) = get_identifier (name_buf);
 
   build_init_ctor (gcov_info_type);
+  build_gcov_exit_decl ();
 
   return true;
 }
diff --git a/gcc/testsuite/g++.dg/gcov/pr16855.C b/gcc/testsuite/g++.dg/gcov/pr16855.C
new file mode 100644
index 000..ec72e99
--- /dev/null
+++ b/gcc/testsuite/g++.dg/gcov/pr16855.C
@@ -0,0 +1,37 @@
+/* { dg-options "-fprofile-arcs -ftest-coverage" } */
+/* { dg-do run { target native } } */
+
+#include 
+using namespace std;
+class Test {
+public:
+	Test(void){
+	cout<< "In Test ctor" << endl;			  /* count(1) */
+	}
+	~Test(void){
+	cout<< "In Test dtor" << endl;			  /* count(1) */
+	}
+}T1;
+
+void uncalled(void){
+	cout<< "In uncalled" << endl;			  /* count(#) */
+}
+int main(void){
+// Test T2;
+cout<< "In main" << endl;  /* count(1) */
+return 0;
+}
+
+#include 
+
+__attribute__((constructor))
+static void construct_navigationBarImages() {
+  fprintf (stderr,  "((construct_navigationBarImages))"); /* count(1) */
+}
+
+__attribute__((destructor))
+static void destroy_navigationBarImages() {
+  fprintf (stderr,  "((destroy_navigationBarImages))");	  /* count(1) */
+}
+
+/* { dg-final { run-gcov branches { -b pr16855.C } } } */
diff --git a/libgcc/libgcov-driver.c b/libgcc/libgcov-driver.c
index d51397e..84471bd 100644
--- a/libgcc/libgcov-driver.c
+++ b/libgcc/libgcov-driver.c
@@ -872,8 +872,8 @@ struct gcov_root __gcov_root;
 struct gcov_master __gcov_master = 
   {GCOV_VERSION, 0};
 
-static void
-gcov_exit (void)
+void
+__gcov_exit (void)
 {
   __gcov_dump_one (&__gcov_root);
   if (__gcov_root.next)
@@ -906,7 +906,6 @@ __gcov_init