[Bug c/11841] New: The code compiled with -funroll-loops crashes

[Bug c/11841] New: The code compiled with -funroll-loops crashes

[panov at canopus dot iacp dot dvo dot ru]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

           Summary: The code compiled with -funroll-loops crashes
           Product: gcc
           Version: 3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: panov at canopus dot iacp dot dvo dot ru
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i486-slackware-linux
  GCC host triplet: i486-slackware-linux
GCC target triplet: i486-slackware-linux

The c code compiled with -funroll-loops and -mcpu=pentium4 crashes. Without
-funroll-loop the program works correctly. gcc-3.2.3 also compiles correct binary.
The program was thoroughly tested with previous versions of gcc and with icc.

[Bug c/11841] The code compiled with -funroll-loops crashes

[panov at canopus dot iacp dot dvo dot ru]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841



------- Additional Comments From panov at canopus dot iacp dot dvo dot ru  2003-08-07 01:24 -------
Created an attachment (id=4575)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=4575&action=view)
crashing program

The program requires GSL library. To create test binary neural2 do "make run".

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[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=11841


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
          Component|c                           |optimization
           Keywords|                            |wrong-code
           Priority|P2                          |P1
            Summary|The code compiled with -    |[3.3 Regression] The code
                   |funroll-loops crashes       |compiled with -funroll-loops
                   |                            |crashes
   Target Milestone|3.4                         |3.3.2


------- Additional Comments From pinskia at physics dot uc dot edu  2003-08-07 01:54 -------
On the mainline (20030806), this works but in 3.3.1 (20030707), it does not and in 3.2.3 it 
works.  I have not tried to reduce it yet.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[steven 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=11841


steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2003-08-07 12:42:57
               date|                            |

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[bangerth at dealii 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=11841


bangerth at dealii dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |steven at gcc dot gnu dot
                   |                            |org


------- Additional Comments From bangerth at dealii dot org  2003-08-07 13:13 -------
Steven, if you place a PR into "confirmed" state, please only do so if you
have a reduced testcase. The audit trail has no indication whatsoever if
you really "analyzed" this case, i.e. whether you can say that the code
is indeed valid, and we don't yet have a small testcase. Would you mind
to provide this?

Thanks
  W.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[s dot bosscher at student dot tudelft dot nl]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841



------- Additional Comments From s dot bosscher at student dot tudelft dot nl  2003-08-07 13:16 -------
Subject: Re:  [3.3 Regression] The code compiled
	with -funroll-loops crashes

Op do 07-08-2003, om 15:13 schreef bangerth at dealii dot org:

> Steven, if you place a PR into "confirmed" state, please only do so if you
> have a reduced testcase. The audit trail has no indication whatsoever if
> you really "analyzed" this case, i.e. whether you can say that the code
> is indeed valid, and we don't yet have a small testcase. Would you mind
> to provide this?


Huh, I didn't confirm any bug???  I just picked up a gfortran bug and
added a comment to another.  Dunno what happend, but it was't me.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[steven 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=11841


steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|steven at gcc dot gnu dot   |
                   |org                         |


------- Additional Comments From steven at gcc dot gnu dot org  2003-08-07 13:18 -------
How did I end up here?

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[bangerth at dealii 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=11841


bangerth at dealii dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |steven at gcc dot gnu dot
                   |                            |org


------- Additional Comments From bangerth at dealii dot org  2003-08-07 13:36 -------
I added you to the list of CCs, that's how you got there. I'll add
you again this time just to make sure you get this mail, but
feel free to remove yourself again :-)

Regarding the confirmation of a PR-- You did this in this message:
  http://gcc.gnu.org/ml/gcc-bugs/2003-08/msg00996.html
This doesn't conform with our policies of only changing the state
of a PR upon confirmation if and when a reduced testcase is
available. There is none for this PR. That's why I objected.

Besides: this definitely is not a fortran bug. Did you by accident
pick the wrong PR?

W.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[s dot bosscher at student dot tudelft dot nl]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841



------- Additional Comments From s dot bosscher at student dot tudelft dot nl  2003-08-07 13:42 -------
Subject: Re:  [3.3 Regression] The code compiled
	with -funroll-loops crashes

Op do 07-08-2003, om 15:36 schreef bangerth at dealii 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=11841
> 
> 
> bangerth at dealii dot org changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |steven at gcc dot gnu dot
>                    |                            |org
> 
> 
> ------- Additional Comments From bangerth at dealii dot org  2003-08-07 13:36 -------
> I added you to the list of CCs, that's how you got there. I'll add
> you again this time just to make sure you get this mail, but
> feel free to remove yourself again :-)
> 
> Regarding the confirmation of a PR-- You did this in this message:
>   http://gcc.gnu.org/ml/gcc-bugs/2003-08/msg00996.html
> This doesn't conform with our policies of only changing the state
> of a PR upon confirmation if and when a reduced testcase is
> available. There is none for this PR. That's why I objected.
> 
> Besides: this definitely is not a fortran bug. Did you by accident
> pick the wrong PR?

I think I know what happened.  I picked up a gfortran bug, and then
wanted to confirm it -- but unlike GNATS, Bugzilla moves to the next bug
in the list when you commit something.  So while I thought I confirmed
the gfortran bug, I actually confirmed this PR by accident.  Sorry about
that.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[bangerth at dealii 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=11841


bangerth at dealii dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|steven at gcc dot gnu dot   |
                   |org                         |
             Status|NEW                         |RESOLVED
         Resolution|                            |INVALID


------- Additional Comments From bangerth at dealii dot org  2003-08-07 13:46 -------
OK, never mind. Let's move this back...

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[bangerth at dealii 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=11841


bangerth at dealii dot org changed:

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


------- Additional Comments From bangerth at dealii dot org  2003-08-07 13:47 -------
...into virginial state.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[ebotcazou 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=11841


ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |WAITING


------- Additional Comments From ebotcazou at gcc dot gnu dot org  2003-09-07 16:33 -------
We need a preprocessed testcase in order to investigate. See
http://gcc.gnu.org/bugs.html

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[pinskia 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=11841



------- Additional Comments From pinskia at gcc dot gnu dot org  2003-09-28 21:33 -------
Most likely the same bug as PR 12340 which is fixed in 3.3.2, can you try a snapshot?

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[panov at canopus dot iacp dot dvo dot ru]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841



------- Additional Comments From panov at canopus dot iacp dot dvo dot ru  2003-10-08 23:26 -------
I have checked with gcc-3.3-20031006, it still creates crashing code with mixing
 -mcpu=pentium4 -funroll-loops. When -mcpu=pentium -funroll-loops (or -mcpu !=
pentium 4) the program does not crash but hangs.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[mmitchel 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=11841


mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|                            |WORKSFORME
   Target Milestone|3.3.2                       |3.3.3


------- Additional Comments From mmitchel at gcc dot gnu dot org  2003-10-16 17:13 -------
Postponed until GCC 3.3.3.

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[panov at canopus dot iacp dot dvo dot ru]

------- Additional Comments From panov at canopus dot iacp dot dvo dot ru  2003-11-18 03:06 -------
I tryed the gcc-3.3-20031117, the bug still arises when "-O2 -mcpu=pentium4
-funroll-loops" used. The program crases (checked with gdb) at the following code:

    for (j = 0; j < GRID_SIZE; j++) {
      for (k = 0; k < GRID_SIZE; k++) {
#ifdef WEIGHTS /* undefined */
        wts_d[j][k] = (double) rand () / max_rand * .5 + .5;
        wts_h[j][k] = (double) rand () / max_rand * .5 + .5;
        wts_v[j][k] = (double) rand () / max_rand * .5 + .5;
#else 
        wts_d[j][k] = 1.0;
        wts_h[j][k] = 1.0; /* This line crashes */
        wts_v[j][k] = 1.0;
#endif 
      }
    }

The compiled program is right when -O1 used. If you use  -mcpu != pentium4 (or
undefined) the program hangs (eats all CPU time). 

My cpu:
 cat /proc/cpuinfo 
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Pentium(R) 4 CPU 2.66GHz
stepping        : 7
cpu MHz         : 2672.776
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov
pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm
bogomips        : 5334.63

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2003-11-18 04:33 -------
Note this was accidentally closed and nobody noticed, woops.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |NEW


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2003-12-12 08:11 -------
We need a preprocessed testcase in order to investigate. See
http://gcc.gnu.org/bugs.html

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[dhazeghi at yahoo dot com]

------- Additional Comments From dhazeghi at yahoo dot com  2004-01-23 18:21 -------
Reopening...

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[dhazeghi at yahoo dot com]

------- Additional Comments From dhazeghi at yahoo dot com  2004-01-23 18:21 -------
Should not be in waiting...

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[gdr at gcc dot gnu dot org]

------- Additional Comments From gdr at gcc dot gnu dot org  2004-02-15 12:35 -------
Adjust milestone

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.3.3                       |3.3.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-03-30 21:38 -------
Here's a reduced testcase:

=========================================
int main()
{
    double d[2];
    int i, j;

    for (i=0; i<2; ++i)
      for (j=0; j<2; ++j)
	d[i+j] = 0;

    return (int)&i;
}
=========================================

This program segfaults when I compile it with "gcc -O2 -funroll-loops"
on my i686-pc-linux-gnu box.
It runs fine, when compiles without "-funroll-loops" or with "-O" instead
of "-O2".


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-03-30 21:43 -------
Hm, Volker, you access d[1+1]=d[2] when d has only two elements. Was  
this intentional, meaning that you want to make the point that only with  
certain flags we segfault, or was this an oversight?  
  
W.  

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |NEW


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-30 21:44 -------
Note that 3.4.0 and 3.5.0 pass even with -fold-unroll-loops which is the -funroll-loops that is in 
the 3.3.x series.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[reichelt at igpm dot rwth-aachen dot de]

------- Additional Comments From reichelt at igpm dot rwth-aachen dot de  2004-03-30 21:57 -------
Subject: Re:  [3.3 Regression] The code compiled with
 -funroll-loops crashes

On 30 Mar, bangerth at dealii dot org wrote:
> 
> ------- Additional Comments From bangerth at dealii dot org  2004-03-30 21:43 -------
> Hm, Volker, you access d[1+1]=d[2] when d has only two elements. Was  
> this intentional, meaning that you want to make the point that only with  
> certain flags we segfault, or was this an oversight?  

Argghh! Stupid mistake on my part. I just went too far reducing the
testcase. But here's a corrected one: It *hangs* when compiled with
-O2 -funroll-loops:

=================================================
int main()
{
    double d[6];
    int i, j;

    for (i=0; i<4; ++i)
      for (j=0; j<3; ++j)
	d[i+j] = 0;

    return (int)&i;
}
=================================================

Sorry for the confusion.




-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-03-30 23:13 -------
My previous testcase only fails with 3.3.x for x>=2.
The following testcase fails (segfaults) with all versions x>=0:

===========================================
int main (int argc, char *argv[])
{
    double d[6];
    int i, j, k=1;

    while (k < argc) ++k;

    for (i=0; i<4; ++i)
      for (j=0; j<3; ++j)
	d[i+j] = 0;

    return (int)&i;
}
===========================================


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[gdr at gcc dot gnu dot org]

------- Additional Comments From gdr at gcc dot gnu dot org  2004-04-27 06:23 -------
Long standing regression since 3.3.0.  Fixed in 3.4.0 and 3.5.0.
Will not fix in 3.3.x.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|3.3.4                       |3.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-04-27 11:56 -------
Since the audit trail is a little confusing, here's a summarized version:

When the following code snippet is compiled on the "-O2 -funroll-loops"
on i686-pc-linux-gnu with gcc 3.3.[234], the resulting executable hangs
(which it should not do):

=================================================
int main()
{
    double d[6];
    int i, j;

    for (i=0; i<4; ++i)
      for (j=0; j<3; ++j)
        d[i+j] = 0;

    return (int)&i;
}
=================================================

A slightly modified version is the following:

===========================================
int main (int argc, char *argv[])
{
    double d[6];
    int i, j, k=1;

    while (k < argc) ++k;

    for (i=0; i<4; ++i)
      for (j=0; j<3; ++j)
        d[i+j] = 0;

    return (int)&i;
}
===========================================

When this code snippet is compiled on the "-O2 -funroll-loops"
on i686-pc-linux-gnu with gcc 3.3 branch, the resulting executable
segfaults (which it should not do). Note, that this example also
misbehaves with gcc 3.3 and 3.3.1.

The bug does not appear in gcc 3.2.3 or gcc 3.4.0 so this is a
"wrong-code 3.3 regression".

Btw, this is the same combination of options that causes PR 13653
to be miscompiled. So they are probably related.

I suspect that there's something wrong with -funroll-loops.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gdr at integrable-solutions
                   |                            |dot net
OtherBugsDependingO|                            |13653
              nThis|                            |
             Status|RESOLVED                    |REOPENED
         Resolution|FIXED                       |
   Target Milestone|3.4.0                       |3.3.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-04-27 14:25 -------
The first (and probably the second, too) testcase can be modified to

int i;
int main()
{
    int d[6];
    int j;

    for (i=0; i<4; ++i)
      for (j=0; j<3; ++j)
        d[i+j] = 0;

    return 0;
}

i.e. remove the weird return.  It was for preventing the variable from being
removed in favor of the biv.  Note that the loop counters need to be exactly
4 and 3 in order to trigger the failure - exchanging the loops also prevents the
failure, as does moving the declaration of i into the body of main.

It is also interesting to note that gcc 3.4 passes the test using
-fold-unroll-loops.  So the loop unroller itself may be not the one to blame.

>From the asm dump (sorry, can't parse RTL) we see:

main:
        pushl   %ebp
        xorl    %eax, %eax
        movl    %esp, %ebp
        subl    $40, %esp
        xorl    %edx, %edx
        andl    $-16, %esp
        movl    %eax, i
.L11:
        movl    $0, -32(%ebp,%edx,4)
        xorl    %eax, %eax
        movl    %eax, -40(%ebp,%edx,4)
        leal    1(%edx), %eax
        xorl    %edx, %edx
        movl    %edx, -40(%ebp,%eax,4)
        leal    2(%ecx), %eax
       ^^^^^^^^^^^^^^^^^^^^^^^^
        xorl    %edx, %edx
        movl    %edx, -40(%ebp,%eax,4)
        xorl    %edx, %edx
        cmpl    $3, %eax
        movl    %edx, -28(%ebp,%ecx,4)
       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        movl    %eax, %edx
        movl    %eax, i
        jle     .L11


the marked line screws the loop counter of the outer loop (in fact - ecx is
never initialized...) and contain a bogous use of it - removing this insn will
fix the loop.

If one looks at the 3.4 version (with -fold-unroll-loops)

main:
        pushl   %ebp
        xorl    %ecx, %ecx
        movl    %esp, %ebp
        subl    $40, %esp
        andl    $-16, %esp
        subl    $16, %esp
        movl    %ecx, i
        .p2align 4,,15
.L9:
        movl    i, %eax
        xorl    %edx, %edx
        xorl    %ecx, %ecx
        movl    %edx, -40(%ebp,%eax,4)
        leal    1(%eax), %edx
        movl    %ecx, -40(%ebp,%edx,4)
        xorl    %ecx, %ecx
        cmpl    $3, %edx
        movl    %ecx, -32(%ebp,%eax,4)
        movl    %edx, i
        jle     .L9

You see a correct version, though strangely enough, still two different
biv's are used for array access (two times %eax and one time %edx). Register
usage could be better, too, as for the -funroll-loops version, which is

main:
        pushl   %ebp
        xorl    %ecx, %ecx
        movl    %esp, %ebp
        subl    $40, %esp
        andl    $-16, %esp
        subl    $16, %esp
        movl    %ecx, i
        .p2align 4,,15
.L9:
        movl    i, %eax
        xorl    %edx, %edx
        xorl    %ecx, %ecx
        movl    %edx, -40(%ebp,%eax,4)
        xorl    %edx, %edx
        movl    %ecx, -36(%ebp,%eax,4)
        movl    %edx, -32(%ebp,%eax,4)
        incl    %eax
        cmpl    $3, %eax
        movl    %eax, i
        jle     .L9

gcc 3.5 and tree-ssa are clever and remove all of main ;)

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-04-27 14:50 -------
A workaround is disabling gcse, or specifying -fno-force-mem, which also makes
the code better:

main:
        pushl   %ebp
        xorl    %ecx, %ecx
        movl    %esp, %ebp
        subl    $40, %esp
        andl    $-16, %esp
        movl    %ecx, i
        .p2align 4,,15
.L11:
        movl    i, %eax
        xorl    %edx, %edx
        xorl    %ecx, %ecx
        movl    %edx, -40(%ebp,%eax,4)
        xorl    %edx, %edx
        movl    %ecx, -36(%ebp,%eax,4)
        movl    %edx, -32(%ebp,%eax,4)
        incl    %eax
        cmpl    $3, %eax
        movl    %eax, i
        jle     .L11


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-04-27 19:32 -------
Investigating.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |ebotcazou at gcc dot gnu dot
                   |dot org                     |org
             Status|REOPENED                    |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-04-27 20:49 -------
Confirmed as a bad interaction between GCSE and the unroller.  Not sure which
one should be fixed.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-04-27 21:20 -------
Subject: Re:  [3.3 Regression] The code compiled with
 -funroll-loops crashes

> Confirmed as a bad interaction between GCSE and the unroller.  Not sure which
> one should be fixed.

Both? ;)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-04-28 08:43 -------
Subject: Re:  [3.3 Regression] The code compiled with
 -funroll-loops crashes

Working around the issue with -fno-force-mem fixes 10 (sic!) failures of
the POOMA testsuite with gcc-3.3 (GCC) 3.3.4 20040301 (prerelease) for me
(other flags are -O2 -funroll-loops).

Even for gcc-3.4.0 there are two extra passes which were previously
segfaults.  A full test for 3.4.0 is running (same options).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-04-28 09:50 -------
Subject: Re:  [3.3 Regression] The code compiled with
 -funroll-loops crashes

Using -fno-force-mem doesn't pessimize generated code for my POOMA
testcases, so maybe we can disable it if -funroll-loops is specified, or
disable it completely by default.

Maybe someone can do a SPEC run with -fno-force-mem/-fforce-mem.  Also in
3.5 we have

2004-03-20  Zdenek Dvorak  <rakdver@atrey.karlin.mff.cuni.cz>

        * emit-rtl.c (set_mem_attributes_minus_bitpos): Set MEM_POINTER
        flag.
        * explow.c (force_not_mem): Set REG_POINTER flag according to
        MEM_POINTER one.
        * rtl.h (MEM_POINTER): New macro.
        (struct rtx_def): Use integrated for MEM_SCALAR_P and frame_related
        for MEM_POINTER.

but I don't know wether this is a bugfix or not.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-04-28 13:30 -------
> Both? ;)

Yeah, but I'm lazy :-)

It turns out that the unroller problem is classic, so I've fixed this one.  But
GCSE has really messed things up.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-04-28 13:42 -------
Subject: Re:  [3.3 Regression] The code compiled with
 -funroll-loops crashes


> It turns out that the unroller problem is classic, so I've fixed this one.  But
> GCSE has really messed things up.

As long as GCSE is not at fault, fine.  But if there is a bug, it should
of course be fixed.  Of course we don't want to work around a bug in the
unroller in GCSE.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-05-12 14:29 -------
Just one more bit of info:

The segfault (second example in comment #27) appeared with the following
gcse-patch by Jan Hubicka:
http://gcc.gnu.org/ml/gcc-cvs/2002-07/msg00472.html

Does reverting this patch help for PR 13653?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-05-12 15:52 -------
Subject: Re:  [3.3 Regression] The code compiled with
 -funroll-loops crashes

On Wed, 12 May 2004, reichelt at gcc dot gnu dot org wrote:

>
> ------- Additional Comments From reichelt at gcc dot gnu dot org  2004-05-12 14:29 -------
> Just one more bit of info:
>
> The segfault (second example in comment #27) appeared with the following
> gcse-patch by Jan Hubicka:
> http://gcc.gnu.org/ml/gcc-cvs/2002-07/msg00472.html
>
> Does reverting this patch help for PR 13653?

Manually reverting the patch doesn't help.

Richard.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug rtl-optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-05-17 21:05 -------
Subject: Bug 11841

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_3-branch
Changes by:	ebotcazou@gcc.gnu.org	2004-05-17 21:05:51

Modified files:
	gcc            : ChangeLog loop.h unroll.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.dg: loop-4.c 

Log message:
	PR optimization/11841
	* loop.h (REGNO_LAST_NOTE_LUID): New macro.
	* unroll.c (unroll_loop): Take into account notes when deciding
	whether a pseudo is local to the loop.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.16114.2.980&r2=1.16114.2.981
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/loop.h.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.65.2.1&r2=1.65.2.2
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/unroll.c.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.184.2.8&r2=1.184.2.9
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.2261.2.374&r2=1.2261.2.375
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/loop-4.c.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=NONE&r2=1.1.4.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841

[Bug rtl-optimization/11841] [3.3 Regression] The code compiled with -funroll-loops crashes

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-05-17 21:09 -------
See http://gcc.gnu.org/ml/gcc-patches/2004-04/msg01895.html


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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11841