http://sourceware.org/bugzilla/show_bug.cgi?id=12944
Summary: GNU LD incorrectly optimize away COMDAT sections
refered from data sections.
Product: binutils
Version: 2.22 (HEAD)
Status: NEW
Severity: normal
Priority: P2
Component: ld
AssignedTo: [email protected]
ReportedBy: [email protected]
Compiling the following testcase:
#include <stdio.h>
#include "x.h"
__attribute__ ((noinline))
void aa(class A *a)
{
a->test();
}
__attribute__ ((noinline))
void bb(class B *b)
{
b->test();
}
int
main(void)
{
class A a;
class B b;
aa (&a);
bb (&b);
}
with LTO and GNU LD:
/abuild/jh/trunk-install/bin/g++ x.cc -O2 -flto -fuse-linker-plugin
-flto-partition=none
Leads to typeinfo being taken away by GNU LD:
00000000004005d2 W _ZTI1A
00000000004005d2 W _ZTI1B
00000000004005d2 W _ZTS1A
00000000004005d2 W _ZTS1B
0000000000400720 r _ZTV1A
0000000000400740 r _ZTV1B
As one can verify the second pointers of _ZTV1A and _ZTV1B are equilvalent
even though they should be different typeinfos:
evans:/abuild/jh/dealIIb/:[0]# gdb a.out
(gdb) p ((void **)0x400720)[0]
$1 = (void *) 0x0
(gdb) p ((void **)0x400720)[1]
$2 = (void *) 0x4005d2
(gdb) p ((void **)0x400720)[2]
$3 = (void *) 0x4006f0
(gdb) p ((void **)0x400740)[0]
$4 = (void *) 0x0
(gdb) p ((void **)0x400740)[1]
$5 = (void *) 0x4005d2
(gdb) p ((void **)0x400740)[2]
$6 = (void *) 0x4006e0
(gdb)
Thinks works well with GOLD:
0000000000400720 V _ZTI1A
0000000000400700 V _ZTI1B
0000000000400730 V _ZTS1A
0000000000400710 V _ZTS1B
00000000004006c0 r _ZTV1A
00000000004006e0 r _ZTV1B
(gdb) p ((void **)0x4006c0)[0]
$1 = (void *) 0x0
(gdb) p ((void **)0x4006c0)[1]
$2 = (void *) 0x400720
(gdb) p ((void **)0x4006c0)[2]
$3 = (void *) 0x400590
(gdb) p ((void **)0x4006e0)[0]
$4 = (void *) 0x0
(gdb) p ((void **)0x4006e0)[1]
$5 = (void *) 0x400700
(gdb) p ((void **)0x4006e0)[2]
$6 = (void *) 0x400580
This is what breaks DealII SPEC2k6 benchmark.
Honza
--
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
_______________________________________________
bug-binutils mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-binutils
