[Bug optimization/11631] New: [3.3-hammer regression] zsh/kernel miscompilation

[Bug optimization/11631] New: [3.3-hammer regression] zsh/kernel miscompilation

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631

           Summary: [3.3-hammer regression] zsh/kernel miscompilation
           Product: gcc
           Version: 3.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: gbeauchesne at mandrakesoft dot com
                CC: aj at suse dot de,gcc-bugs at gcc dot gnu dot org,jh at
                    suse dot cz
  GCC host triplet: i686-pc-linux-gnu

The following program fails at -O2 but is OK at -Os. This occurs on both AMD64 &
 IA-32 but only on 3.3-hammer branch as of today. This is a regression from
3.3-hammer 2003/05/24. I believe this is similar to the kernel miscompilation
noticed by users at -O2 but zsh was easier to hunt. ;-)

Note the specific placement of f() & g(), the use of a static global variable,
the for(;;) loop.

/* Extracted from zsh sources.  */
/* Contributed by Gwenole Beauchesne <gbeauchesne@mandrakesoft.com>  */

extern void abort(void);
extern void exit(int);

static void f(const char *name);
static void g(const char *name);

static char *p;

int main(void)
{
  p = "foo";
  do {
      char *arg0 = p;
      if (!(p = strrchr(arg0, '/')))
	  p = arg0;
      else
	  p++;
      if (strcmp(p, "bar") != 0)
		break;
  } while (p);
  
  f(p);
  g(p);

  for (;;) {
  }

  return 0;
}

static void f(const char *name)
{
}

static void g(const char *name)
{
  if (name)
	exit(0);
  else
	abort();
}

[Bug optimization/11631] [3.3-hammer regression] zsh/kernel miscompilation

[matz at suse dot de]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From matz at suse dot de  2003-07-22 15:57 -------
Subject: Re:  New: [3.3-hammer regression] zsh/kernel
 miscompilation

> The following program fails at -O2 but is OK at -Os.

In which sense does it fail?  It aborts()?  I ask because I can't
reproduce it with a clean hammer branch checkout (i.e. no local SuSE
patches):

% cvs -z2 -q up
% cd ../devel
% make
... make output ...
% ./gcc/xgcc -B./gcc/ -O2 pr11631.c
% ./a.out
% echo $?
0

Also some other flag combinations (-O3, -fPIC) didn't make it fail.
I admit that this isn't with a bootstrapped compiler, but instead just
with a simple 'make''d one.  I tested this on i686-pc-linux-gnu.  Could
you verify?

[Bug optimization/11631] [3.3-hammer regression] zsh/kernel miscompilation

[falk at debian dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From falk at debian dot org  2003-07-22 16:04 -------
I can reproduce it with a one week old mainline compiler on alphaev68-linux.
Maybe it's been fixed meanwhile, I can't test because gcc doesn't bootstrap
for me currently because of weird problems with gperf.

falk@juist:~% gcc --version
gcc (GCC) 3.4 20030714 (experimental)
falk@juist:~% gcc 11631.c && ./a.out
falk@juist:~% gcc 11631.c -O2 && ./a.out
zsh: abort (core dumped)  ./a.out

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631


gbeauchesne at mandrakesoft dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[3.3-hammer regression]     |[3.4/3.3-hammer regression]
                   |zsh/kernel miscompilation   |zsh/kernel miscompilation


------- Additional Comments From gbeauchesne at mandrakesoft dot com  2003-07-22 16:06 -------
To be more specific, this fails with:
- GNU C version 3.4 20030722 (experimental) (i686-pc-linux-gnu)
- GNU C version 3.3.1-hammer 20030703 (prerelease) (i686-pc-linux-gnu)
[also on AMD64 for 3.3-hammer]

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From gbeauchesne at mandrakesoft dot com  2003-07-22 16:43 -------
Subject: Re:  [3.3-hammer regression] zsh/kernel
 miscompilation

On Tue, 22 Jul 2003, matz at suse dot de wrote:

> > The following program fails at -O2 but is OK at -Os.
> 
> In which sense does it fail?  It aborts()?

Yes.

>   I ask because I can't reproduce it with a clean hammer branch checkout
> (i.e. no local SuSE patches):

Likewise to here. But with recent 3.3-hammer you have to add 
-fno-unit-at-a-time since that's enabled on the branch.

Just ./contrib/gcc_update -d as of today:

[gb@hp6 gcc]$ ./xgcc -B $PWD ~/vrac/zsh-bug.c -O2 -fno-unit-at-a-time
[gb@hp6 gcc]$ ./a.out
Aborted

GNU C version 3.3.1-hammer 20030703 (prerelease) (i686-pc-linux-gnu)
actually refered to co from 20030708, according to ChangeLog.hammer.

Re: [Bug optimization/11631] [3.3-hammer regression] zsh/kernel miscompilation

[Gwenole Beauchesne]

On Tue, 22 Jul 2003, matz at suse dot de wrote:

> > The following program fails at -O2 but is OK at -Os.
> 
> In which sense does it fail?  It aborts()?

Yes.

>   I ask because I can't reproduce it with a clean hammer branch checkout
> (i.e. no local SuSE patches):

Likewise to here. But with recent 3.3-hammer you have to add 
-fno-unit-at-a-time since that's enabled on the branch.

Just ./contrib/gcc_update -d as of today:

[gb@hp6 gcc]$ ./xgcc -B $PWD ~/vrac/zsh-bug.c -O2 -fno-unit-at-a-time
[gb@hp6 gcc]$ ./a.out
Aborted

GNU C version 3.3.1-hammer 20030703 (prerelease) (i686-pc-linux-gnu)
actually refered to co from 20030708, according to ChangeLog.hammer.

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[zlomek at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631


zlomek at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2003-07-24 09:08:51
               date|                            |


------- Additional Comments From zlomek at gcc dot gnu dot org  2003-07-24 09:08 -------
cc1 -O2 zsh.c     # does not work
cc1 -O2 -fno-gcse zsh.c  # works

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-24 13:26 -------
It also happens on powerpc-apple-darwin6.6 on the mainline (20030724).

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[zlomek at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631


zlomek at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rakdver at atrey dot karlin
                   |                            |dot mff dot cuni dot cz
           Keywords|                            |wrong-code


------- Additional Comments From zlomek at gcc dot gnu dot org  2003-07-24 16:14 -------
it is a bug in store motion,
when compiled with "./xgcc -B . -O2 -fno-gcse-sm zsh.c" it works.

BB 1 (see testcase):                          0                                
                                                                               
     
char *arg0 = p;                               |                                
                                                                               
     
if ((p = strrchr(arg0, '/')))                 1 <<                             
                                                                               
     
goto BB3                                     /|  ^                             
                                                                               
     
                                            2 |  ^                             
                                                                                    
BB 2:                                        \|  ^                             
                                                                                   
p = arg0;                                     3 >^                             
                                                                                   
                                              |                                
                                                                               
     
BB 3:                                         4                                
                                                                               
     
if (strcmp(p, "bar") == 0)                                                     
                                                                               
     
 goto BB1;                                                                     
                                                                               
     
                                                                               
                                                                               
     
BB 4:                                                                          
                                                                               
     
  f(p);                                                                        
                                                                               
     
  g(p);                                                                        
                                                                               
     
                                                                               
                                                                               
     
The store to p in BB1 is replaced by store to a pseudoreg.                     
                                                                               
       
In front of BB4 the pseudoreg is stored to p.                                  
                                                                               
     
The problem is that when we store to p in BB 2                                 
                                                                               
     
it will be replaced in BB 4 by the pseudoreg (result of strrchr, which is 0).  
                                                                               
                          
I think the store to p in BB 2 should be replaced with a store to the same     
                                                                               
     
pseudoreg too.

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From gbeauchesne at mandrakesoft dot com  2003-07-24 17:35 -------
Subject: Re:  [3.4/3.3-hammer regression] zsh/kernel
 miscompilation

On Thu, 24 Jul 2003, zlomek at gcc dot gnu dot org wrote:

> it is a bug in store motion,

Indeed, this coincidates with the following patch merged from mainline to 
3.3-hammer branch between 2003/05/20 & 2003/07/08.

The reduced testcase also fails on current 3.3-hammer branch with -O2 
-fno-unit-at-a-time.

Store motion got enabled with this patch:
<http://gcc.gnu.org/ml/gcc-patches/2003-02/msg02090.html>

2003-06-21  Zdenek Dvorak  <rakdver@atrey.karlin.mff.cuni.cz>

	Merge from mainline:
		* gcse.c (struct ls_expr): Added pattern_regs field.
		(ldst_entry): Initialize it.
		(extract_mentioned_regs, extract_mentioned_regs_helper): 
New.
		(store_ops_ok): Use regs precomputed by them.
		(find_loads, store_killed_in_insn, load_kills_store): 
Change return
		type to bool.
		(store_killed_before, store_killed_after): Take position 
of register
		set in account.
		(reg_set_info): Store position of the setter.
		(gcse_main): Enable store motion.
		(mems_conflict_for_gcse_p): Enable load motion of 
non-symbol mems.
		(pre_insert_copy_insn, update_ld_motion_stores, 
insert_store): Prevent rtl
		sharing.
		(simple_mem): Enable store motion of non-symbol mems.
		(regvec): Type changed.
		(LAST_AVAIL_CHECK_FAILURE): New.
		(compute_store_table_current_insn): New.
		(build_store_vectors): Computation of availability and 
anticipatability
		moved ...
		(compute_store_table, find_moveable_store): ... here.
		(delete_store): Remove senseless comment.
		(store_motion): Reorganize.
	* gcse.c (local_variable_p): New.

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[rakdver at atrey dot karlin dot mff dot cuni dot cz]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From rakdver at atrey dot karlin dot mff dot cuni dot cz  2003-07-24 18:49 -------
Subject: Re:  [3.4/3.3-hammer regression] zsh/kernel miscompilation

The problem is with the infinite loop.  I am not really sure where
things get wrong (it happens inside lcm.c, which I take as a blackbox),
but this patch fixes it.  I will commit it as obvious unless it breaks
something else.

Zdenek

Changelog:
	* gcse.c (store_motion): Connect infinite loops to exit.

Index: gcse.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/gcse.c,v
retrieving revision 1.222.2.11
diff -c -3 -p -r1.222.2.11 gcse.c
*** gcse.c	25 Jun 2003 20:22:48 -0000	1.222.2.11
--- gcse.c	24 Jul 2003 18:44:54 -0000
*************** store_motion ()
*** 7681,7686 ****
--- 7681,7687 ----
    /* Now compute kill & transp vectors.  */
    build_store_vectors ();
    add_noreturn_fake_exit_edges ();
+   connect_infinite_loops_to_exit ();
  
    edge_list = pre_edge_rev_lcm (gcse_file, num_stores, transp, ae_gen,
  				st_antloc, ae_kill, &pre_insert_map,

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[aj at suse dot de]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From aj at suse dot de  2003-07-24 19:13 -------
Subject: Re:  [3.4/3.3-hammer regression] zsh/kernel
 miscompilation

Zdenek Dvorak <rakdver@atrey.karlin.mff.cuni.cz> writes:

> The problem is with the infinite loop.  I am not really sure where
> things get wrong (it happens inside lcm.c, which I take as a blackbox),
> but this patch fixes it.  I will commit it as obvious unless it breaks
> something else.
>
> Zdenek
>
> Changelog:

PLease mention in the changelog the bug like this:

        PR optimization/11631 
 	* gcse.c (store_motion): Connect infinite loops to exit.

thanks,
Andreas

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[cvs-commit at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From cvs-commit at gcc dot gnu dot org  2003-07-24 20:59 -------
Subject: Bug 11631

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rakdver@gcc.gnu.org	2003-07-24 20:59:29

Modified files:
	gcc            : ChangeLog gcse.c 

Log message:
	PR optimization/11631
	* gcse.c (store_motion): Connect infinite loops to exit.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.616&r2=2.617
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/gcse.c.diff?cvsroot=gcc&r1=1.262&r2=1.263

[Bug optimization/11631] [3.4/3.3-hammer regression] zsh/kernel miscompilation

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631


pinskia at physics dot uc dot edu changed:

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


------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-24 21:02 -------
Fixed by the above patch.

[Bug optimization/11631] [3.3-hammer regression] zsh/kernel miscompilation

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|FIXED                       |
            Summary|[3.4/3.3-hammer regression] |[3.3-hammer regression]
                   |zsh/kernel miscompilation   |zsh/kernel miscompilation


------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-24 21:04 -------
Keep it open until the branch 3.3-hammer gets fixed.

[Bug optimization/11631] [3.3-hammer regression] zsh miscompilation

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631


gbeauchesne at mandrakesoft dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[3.3-hammer regression]     |[3.3-hammer regression] zsh
                   |zsh/kernel miscompilation   |miscompilation


------- Additional Comments From gbeauchesne at mandrakesoft dot com  2003-07-25 08:36 -------
I haven't tried the patch yet but simply disabled store_motion() optimization.
That "fixed" zsh miscompilation but not kernel. Investigating this one. Thusly
removing kernel from summary.

[Bug optimization/11631] [3.3-hammer regression] zsh miscompilation

[cvs-commit at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631



------- Additional Comments From cvs-commit at gcc dot gnu dot org  2003-07-28 19:17 -------
Subject: Bug 11631

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	hammer-3_3-branch
Changes by:	rakdver@gcc.gnu.org	2003-07-28 19:17:54

Modified files:
	gcc            : gcse.c ChangeLog.hammer 

Log message:
	PR optimization/11631
	* gcse.c (store_motion): Connect infinite loops to exit.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/gcse.c.diff?cvsroot=gcc&only_with_tag=hammer-3_3-branch&r1=1.222.2.11&r2=1.222.2.12
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.hammer.diff?cvsroot=gcc&only_with_tag=hammer-3_3-branch&r1=1.1.2.157&r2=1.1.2.158

[Bug optimization/11631] [3.3-hammer regression] zsh miscompilation

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11631


pinskia at physics dot uc dot edu changed:

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


------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-28 19:20 -------
Fixed in 3.4 and now on the 3.3-hammer branch.