[Bug c/11399] New: gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[Bug c/11399] New: gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[ianw at gelato dot unsw dot edu dot au]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11399

           Summary: gcc 3.3.1 miscompiles an if - else if - else statement
                    with O1 on ia64
           Product: gcc
           Version: 3.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: ianw at gelato dot unsw dot edu dot au
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: ia64-unknown-linux-gnu
  GCC host triplet: ia64-unknown-linux-gnu
GCC target triplet: ia64-unknown-linux-gnu

gcc 3.3.1 miscompiles the following program at optimisation >= 1 on ia64.  The
program will segfault.  I noticed this when I was building Apache.


--- start test.c ---
#include <stdio.h>
#include <string.h>
 
int retval;
 
const char *if_else(void)
{
 
    if (!strcasecmp("this is", "false")) {
        retval = 0;
    } else if (!strcasecmp("true", "true")) {
        retval = 1;
    } else
        return "a string";
 
    return "ok";
}
 
int main(void)
{
   printf("%s\n", if_else());
}

--- end test.c ---

The offending code from gcc -S

        br.call.sptk.many b0 = strcasecmp#
        ;;
        .mii
        mov r1 = r34
        cmp4.ne p6, p7 = 0, r8
        ;;
        addl r14 = @ltoffx(retval#), r1
        ;;
        .mii
        ld8.mov r15 = [r14], retval#
        addl r14 = @ltoffx("a string"), r1
        ;;
        (p7) addl r14 = 1, r0               <---- set r14 to 1
        ;;
        .mmb
        (p6) ld8.mov r8 = [r14], "a string" <---- segfault
        (p7) st4 [r15] = r14
        (p7) br.cond.dptk .L3

This does compile OK with gcc 3.2 and gcc 3.4

[Bug c/11399] gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[ianw at gelato dot unsw dot edu dot au]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11399



------- Additional Comments From ianw at gelato dot unsw dot edu dot au  2003-07-02 00:10 -------
Created an attachment (id=4319)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=4319&action=view)
test case

[Bug c/11399] gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[ianw at gelato dot unsw dot edu dot au]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11399



------- Additional Comments From ianw at gelato dot unsw dot edu dot au  2003-07-02 00:11 -------
Created an attachment (id=4320)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=4320&action=view)
output from gcc -S 3.3.1

[Bug c/11399] gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[ianw at gelato dot unsw dot edu dot au]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11399



------- Additional Comments From ianw at gelato dot unsw dot edu dot au  2003-07-02 00:12 -------
Created an attachment (id=4321)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=4321&action=view)
correct output from gcc -S 3.4

[Bug c/11399] gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[ianw at gelato dot unsw dot edu dot au]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11399


ianw at gelato dot unsw dot edu dot au changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Attachment #4320|application/octet-stream    |text/plain
          mime type|                            |

[Bug c/11399] gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[wilson at tuliptree 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=11399



------- Additional Comments From wilson at tuliptree dot org  2003-07-04 05:07 -------
Subject: Re:  New: gcc 3.3.1 miscompiles an if - else if - else
 statement with O1 on ia64

ianw at gelato dot unsw dot edu dot au wrote:
>         cmp4.ne p6, p7 = 0, r8
>         addl r14 = @ltoffx("a string"), r1
>         ;;
>         (p7) addl r14 = 1, r0               <---- set r14 to 1
>         ;;
>         .mmb
>         (p6) ld8.mov r8 = [r14], "a string" <---- segfault

You are misreading the code.  IPF has predicated execution.  An 
instruction executes only if the qualifying predicate is true.  In this 
case, the two instructions use different predicate registers p6/p7, and 
because this is a predicate register pair we are guaranteed that at most 
one is true, i.e. they are mutex in IPF terminology.

The 3.3 and 3.4 .s files you put in the bug report don't match your 
annotated assembly sample.  The annotated code has p6/p7 predicate 
register usage.  The 3.3/3.4 .s output have only p7 predicate register 
usage.

I am unable to reproduce your assembly output using a x86-x-ia64 cross 
compiler using 3.3 branch last updated a few days ago.

I don't have access to IPF hardware, so I am unfortunately not able to 
run the example.

Jim

[Bug c/11399] gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

[ianw at gelato dot unsw dot edu dot au]

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11399



------- Additional Comments From ianw at gelato dot unsw dot edu dot au  2003-07-04 06:31 -------
Subject: Re:  gcc 3.3.1 miscompiles an if - else if - else statement with O1 on ia64

On Fri, Jul 04, 2003 at 05:07:13AM -0000, wilson at tuliptree dot org wrote:
> ------- Additional Comments From wilson at tuliptree dot org  2003-07-04 05:07 -------
> ianw at gelato dot unsw dot edu dot au wrote:
> >         cmp4.ne p6, p7 = 0, r8
> >         addl r14 = @ltoffx("a string"), r1
> >         ;;
> >         (p7) addl r14 = 1, r0               <---- set r14 to 1
> >         ;;
> >         .mmb
> >         (p6) ld8.mov r8 = [r14], "a string" <---- segfault

I pasted in the wrong example -- that p6 in the annoted code was put
in by *me* playing with the output of -S when I found the problem.
Sorry!!  However, the attached examples are verbatim, gcc 3.3.1
produces the code :

	(p7) ld8.mov r15 = [r14], retval#
	(p7) addl r14 = 1, r0
	;;
	(p7) st4 [r15] = r14
	(p7) br.cond.dptk .L3

> The 3.3/3.4 .s output have only p7 predicate register 
> usage.

gcc 3.4 also only uses p7, but it uses r15.

	(p7) addl r15 = 1, r0
	(p7) addl r14 = @ltoffx(retval#), r1
	;;
	(p7) ld8.mov r14 = [r14], retval#
	;;
	(p7) st4 [r14] = r15
	(p7) br.cond.dptk .L3
 
> I am unable to reproduce your assembly output using a x86-x-ia64 cross 
> compiler using 3.3 branch last updated a few days ago.

ok, I just built from CVS gcc-3_3-branch 5 minutes ago :
ianw@tartufi:~/tmp/gcc-bug$ cvs-gcc -v
Reading specs from /usr/local/lib/gcc-lib/ia64-unknown-linux-gnu/3.3.1/specs
Configured with: ./configure --enable-languages=c --program-prefix=cvs- --enable-shared
Thread model: posix
gcc version 3.3.1 20030704 (prerelease)

and it still does it with -O2

> I don't have access to IPF hardware, so I am unfortunately not able to 
> run the example.

If you think it would help I can most probably get you access to some;
please email me privately.

Otherwise, can you suggest where in the code you think the difference
between 3.3 and 3.4 is likely to be?  I'd like to get the fix
backported, but currently I'm a bit lost at where to look.

Thankyou for looking at the bug,

-i

[Bug target/11399] [3.3 regression] gcc miscompiles an if - else if - else statement with -O2 on ia64

[dhazeghi at yahoo 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=11399


dhazeghi at yahoo dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
             Status|UNCONFIRMED                 |NEW
          Component|c                           |target
     Ever Confirmed|                            |1
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2003-07-06 01:00:45
               date|                            |
            Summary|gcc 3.3.1 miscompiles an if |[3.3 regression] gcc
                   |- else if - else statement  |miscompiles an if - else if
                   |with O1 on ia64             |- else statement with -O2 on
                   |                            |ia64


------- Additional Comments From dhazeghi at yahoo dot com  2003-07-06 01:00 -------
Confirmed problem still exists with gcc 3.3 branch (20030705). In the future though, please 
include preprocessed source with your reports though (it makes it possible to see these problems 
with cross-compilers...). This is a regression from gcc 3.2.3.

[Bug target/11399] [3.3 regression] gcc miscompiles an if - else if - else statement with -O2 on ia64

[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=11399


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4                         |3.3.1

[Bug target/11399] [3.3 regression] gcc miscompiles an if - else if - else statement with -O2 on ia64

[wilson at tuliptree 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=11399



------- Additional Comments From wilson at tuliptree dot org  2003-07-07 06:39 -------
Subject: Re:  gcc 3.3.1 miscompiles an if - else if - else statement
 with O1 on ia64

ianw at gelato dot unsw dot edu dot au wrote:
>>I don't have access to IPF hardware, so I am unfortunately not able to 
>>run the example.
> If you think it would help I can most probably get you access to some;
> please email me privately.

Well, it is really more an issue of time than hardware access. I can 
probably get a free machine if I ask, but I am doing so much stuff for 
GCC that I haven't had time to worry about doing IA-64 specific work.

Jim

[Bug target/11399] [3.3 regression] gcc miscompiles an if - else if - else statement with -O2 on ia64

[neroden 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=11399



------- Additional Comments From neroden at gcc dot gnu dot org  2003-07-09 02:35 -------
There've been a ludicrous number of changes in IA64 between 3.3 and 3.4.  Maybe you could check if the bug exists at the 3.3 branchpoint, and if it 
doesn't, a regression hunt can be done on the branch, and if it does, a, uh, "bugfix hunt" can be done on the mainline.

[Bug target/11399] [3.3 regression] gcc miscompiles an if - else if - else statement with -O2 on ia64

[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=11399


steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P2                          |P1


------- Additional Comments From steven at gcc dot gnu dot org  2003-07-11 23:36 -------
A wrong-code regression, so should be P1.

[Bug target/11399] [3.3 regression] gcc miscompiles an if - else if - else statement with -O2 on ia64

[neroden 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=11399



------- Additional Comments From neroden at gcc dot gnu dot org  2003-07-11 23:40 -------
"Priority" is for assigned people to manage their bug lists; don't bother messing with it unless the bug is assigned.  Severity: critical is good 
enough for what you want.

[Bug target/11399] [3.3 regression] gcc miscompiles an if - else if - else statement with -O2 on ia64

[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=11399


ebotcazou at gcc dot gnu dot org changed:

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


------- Additional Comments From ebotcazou at gcc dot gnu dot org  2003-07-12 07:45 -------
Proposed patch:
http://gcc.gnu.org/ml/gcc-patches/2003-07/msg01044.html


*** This bug has been marked as a duplicate of 11320 ***