c/4214: powerpc64: wrong code from optimized mask and shift

c/4214: powerpc64: wrong code from optimized mask and shift

[Kevin Ryde]

>Number:         4214
>Category:       c
>Synopsis:       powerpc64: wrong code from optimized mask and shift
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Mon Sep 03 16:46:00 PDT 2001
>Closed-Date:
>Last-Modified:
>Originator:     
>Release:        3.0
>Organization:
>Environment:
System: AIX panther 3 4 000785504C00
host: powerpc-ibm-aix4.3.2.0
build: powerpc-ibm-aix4.3.2.0
target: powerpc-ibm-aix4.3.2.0
configured with: ../gcc-3.0/configure 

>Description:
A certain mask and shift combination produces a wrong answer when
compiled with optimization.

>How-To-Repeat:
A program foo.c,

        int n = 3;

        int
        main (void)
        {
          printf ("%d\n", (n & 2) << 1);
          return 0;
        }

when compiled with

        gcc -O -maix64 -mpowerpc64 foo.c

and run just as "./a.out", prints 6 whereas I think it should print 4
(and which it does if compiled without the -O).

For what it's worth, it seems this only happens under "-maix64
-mpowerpc64", not in 32-bit mode.
>Fix:
>Release-Note:
>Audit-Trail:
>Unformatted:

Re: c/4214: powerpc64: wrong code from optimized mask and shift

[dje]

Synopsis: powerpc64: wrong code from optimized mask and shift

Responsible-Changed-From-To: unassigned->amodra
Responsible-Changed-By: dje
Responsible-Changed-When: Mon Sep 24 14:52:54 2001
Responsible-Changed-Why:
    PowerPC
State-Changed-From-To: open->closed
State-Changed-By: dje
State-Changed-When: Mon Sep 24 14:52:54 2001
State-Changed-Why:
    Fixed in gcc-3.0.2 and gcc-3.1

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=4214&database=gcc

Re: c/4214: powerpc64: wrong code from optimized mask and shift

[Mark Mitchell]

The following reply was made to PR c/4214; it has been noted by GNATS.

From: Mark Mitchell <mark@codesourcery.com>
To: Alan Modra <amodra@bigpond.net.au>, Kevin Ryde <user42@zip.com.au>
Cc: "gcc-gnats@gcc.gnu.org" <gcc-gnats@gcc.gnu.org>,
        David Edelsohn <dje@watson.ibm.com>
Subject: Re: c/4214: powerpc64: wrong code from optimized mask and shift
Date: Mon, 03 Sep 2001 19:56:41 -0700

 > Is the patch OK for the branch Mark?
 
 It's fine by me, but I'd like David's thoughts.  David, unless this
 patch makes you nervous, would you approve it?
 
 --
 Mark Mitchell                   mark@codesourcery.com
 CodeSourcery, LLC               http://www.codesourcery.com
 

Re: c/4214: powerpc64: wrong code from optimized mask and shift

[Alan Modra]

The following reply was made to PR c/4214; it has been noted by GNATS.

From: Alan Modra <amodra@bigpond.net.au>
To: Kevin Ryde <user42@zip.com.au>, mark@codesourcery.com
Cc: gcc-gnats@gcc.gnu.org, David Edelsohn <dje@watson.ibm.com>
Subject: Re: c/4214: powerpc64: wrong code from optimized mask and shift
Date: Tue, 4 Sep 2001 11:51:06 +0930

 On Tue, Sep 04, 2001 at 09:39:45AM +1000, Kevin Ryde wrote:
 > 
 > A certain mask and shift combination produces a wrong answer when
 > compiled with optimization.
 > 
 > >How-To-Repeat:
 > A program foo.c,
 > 
 >         int n = 3;
 > 
 >         int
 >         main (void)
 >         {
 >           printf ("%d\n", (n & 2) << 1);
 >           return 0;
 >         }
 
 Fixed by http://gcc.gnu.org/ml/gcc-patches/2001-08/msg01493.html ,
 which I applied to mainline.
 
 Is the patch OK for the branch Mark?  You could argue that it contains
 optimisations (extra patterns for rldicr, no bit-shifting loops) as well
 as the bugfix for rldic, but I'm far more likely to make an error if I
 try to split the patch rather than apply the lot.  ;-)
 
 Alan