[Bug c/11319] [3.3 regression] loop miscompiled on ppc32

[Bug c/11319] [3.3 regression] loop miscompiled on ppc32

[marcus at jet dot franken 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=11319



------- Additional Comments From marcus at jet dot franken dot de  2003-06-25 13:21 -------
Created an attachment (id=4279)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=4279&action=view)
testcase for 11319

[Bug c/11319] New: [3.3 regression] loop miscompiled on ppc32

[marcus at jet dot franken 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=11319

           Summary: [3.3 regression] loop miscompiled on ppc32
           Product: gcc
           Version: 3.3
            Status: UNCONFIRMED
          Severity: critical
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: marcus at jet dot franken dot de
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: powerpc-unknown-linux-gnu
  GCC host triplet: powerpc-unknown-linux-gnu
GCC target triplet: powerpc-unknown-linux-gnu

Hi,  
  
attached testcase, extracted from the cast-256.c crypto module in libmcrypt,   
miscompiles on ppc32 with 3.2, 3.3 branch (gcc version 3.3.1 20030625 
(prerelease)) and HEAD (gcc version 3.4 20030625 (experimental)) with -O2:  
 
$ gcc -o xx ~meissner/cast-256.c  -O2  
$ ./xx  
xx: xx.c:27: main: Assertion `a == 0x13' failed.  
$  
  
With -fno-strength-reduce it starts to work:  
$ gcc -o xx ~meissner/cast-256.c  -O2 -fno-strength-reduce  
$ ./xx

[Bug c/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |DUPLICATE
            Summary|[3.3 regression] loop       |[3.3/3.4 regression] loop
                   |miscompiled on ppc32        |miscompiled on ppc32


------- Additional Comments From pinskia at physics dot uc dot edu  2003-06-25 13:35 -------
This is a total dup of 9745 but I will put the simplified testcase in that one.

*** This bug has been marked as a duplicate of 9745 ***

[Bug c/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[wilson 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=11319


wilson at gcc dot gnu dot org changed:

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


------- Additional Comments From wilson at gcc dot gnu dot org  2003-07-11 06:52 -------
The patch that fixed PR 9745 did not fix this one. It is a closely related
problem, but there are multiple problems here. See
http://gcc.gnu.org/ml/gcc/2003-07/msg00668.html
for a discussion of various issues related to this problem.

[Bug c/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319


dhazeghi at yahoo dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2003-07-12 06:17:07
               date|                            |


------- Additional Comments From dhazeghi at yahoo dot com  2003-07-12 06:17 -------
Confirmed with gcc 3.3 branch and mainline (20030710).

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c                           |optimization
  GCC build triplet|powerpc-unknown-linux-gnu   |powerpc-*-*
   GCC host triplet|powerpc-unknown-linux-gnu   |powerpc-*-*
 GCC target triplet|powerpc-unknown-linux-gnu   |powerpc-*-*


------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-15 13:58 -------
I could also reproduce this on powerpc-apple-darwin6.6 so this is at least powerpc-*-*.

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4                         |3.3.2

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[janis187 at us dot ibm 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=11319


janis187 at us dot ibm dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rth at redhat dot com


------- Additional Comments From janis187 at us dot ibm dot com  2003-08-05 21:15 -------
The regression in PR 11319 was introduced or exposed by this patch:

2002-02-07  Richard Henderson  <rth@redhat.com>

        PR optimization/2463
        * alias.c (find_base_value): Recall base values for fixed hard regs.
        * loop.c (loop_regs_update): Don't use single_set on non-insns.

The regression hunt took place on powerpc-linux using the test case from
the attachment in comment #1 compiled with -O2 and then executed.

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[wilson 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=11319



------- Additional Comments From wilson at gcc dot gnu dot org  2003-08-11 23:56 -------
Richard's patch did not introduce the problem, it just exposed it.  In the old
days, a sequence was an insn chain if more than one instruction, and a pattern
if only one instruction.  In the single insn case, which is the common case, the
loop.c code was passing the pattern to single_set which failed, and thus
prevented it from calling record_base_value.  Richard's patch fixed it to call
record_base_value as intended, thus exposing the alias.c problem fixed by my
patch in PR 10021.

This incidentally points out a problem with loop_regs_update.  The !INSN_P code
is obsolete and should be removed.

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319



------- Additional Comments From cvs-commit at gcc dot gnu dot org  2003-08-12 05:26 -------
Subject: Bug 11319

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_3-branch
Changes by:	wilson@gcc.gnu.org	2003-08-12 05:26:15

Modified files:
	gcc            : ChangeLog alias.c 

Log message:
	PR optimization/11319
	PR target/10021
	* alias.c (find_base_value, case REG): Return 0 not src if no base
	found.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.16114.2.706&r2=1.16114.2.707
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/alias.c.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.181.2.2&r2=1.181.2.3

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319



------- Additional Comments From cvs-commit at gcc dot gnu dot org  2003-08-12 05:28 -------
Subject: Bug 11319

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	wilson@gcc.gnu.org	2003-08-12 05:28:45

Modified files:
	gcc            : ChangeLog alias.c 

Log message:
	PR optimization/11319
	PR target/10021
	* alias.c (find_base_value, case REG): Return 0 not src if no base
	found.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.769&r2=2.770
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/alias.c.diff?cvsroot=gcc&r1=1.198&r2=1.199

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[wilson 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=11319


wilson at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zippel at linux-m68k dot org


------- Additional Comments From wilson at gcc dot gnu dot org  2003-08-28 07:36 -------
*** Bug 10021 has been marked as a duplicate of this bug. ***

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |debian-gcc at lists dot
                   |                            |debian dot org


------- Additional Comments From pinskia at gcc dot gnu dot org  2003-09-02 22:46 -------
*** Bug 11366 has been marked as a duplicate of this bug. ***

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[wilson 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=11319



------- Additional Comments From wilson at gcc dot gnu dot org  2003-09-07 05:35 -------
The bug has been fixed by the alias.c patch.  However, there were questions
raised about the effect of the patch on the performance of gcc compiled code.  I
have left the bug report open for now as a reminder to look into this issue.

I have checked in a patch to fix the loop_regs_update !INSN_P problem I
mentioned earlier.
  http://gcc.gnu.org/ml/gcc-patches/2003-09/msg00397.html

I have done SPEC testing on a Pentium4 system running GNU/Linux.  I used -O2
-march=pentium4 as the gcc optimization options.  I see no significant effect on
the performance for those benchmarks that I could get running.  I do see a
noticable drop in 301.apsi performance after the first loop_regs_update patch
went in, but it went back up after the alias.c patch went in, so the problem
seems to have corrected itself.

PowerPC performance tests are pending.

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[wilson 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=11319


wilson at gcc dot gnu dot org changed:

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


------- Additional Comments From wilson at gcc dot gnu dot org  2003-09-11 00:59 -------
I have now run SPEC on a PowerMac G5.  The only noticable performance change I
see is a 3% improvement in mesa with my patches, but 3% is small enough that it
could be within my measurement error.

David Edelsohn has agreed that my patches did not introduce performance
regressions.  See
    http://gcc.gnu.org/ml/gcc/2003-09/msg00377.html

So I am not closing this problem report.

[Bug optimization/11319] [3.3/3.4 regression] loop miscompiled on ppc32

[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=11319


ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |selinger at mathstat dot
                   |                            |uottawa dot ca


------- Additional Comments From ebotcazou at gcc dot gnu dot org  2003-09-11 07:05 -------
*** Bug 12243 has been marked as a duplicate of this bug. ***