Re: optimization/7632: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments

Re: optimization/7632: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments

[dje]

Synopsis: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments

Responsible-Changed-From-To: unassigned->dje
Responsible-Changed-By: dje
Responsible-Changed-When: Wed Sep 11 12:06:03 2002
Responsible-Changed-Why:
    PowerPC
State-Changed-From-To: open->closed
State-Changed-By: dje
State-Changed-When: Wed Sep 11 12:06:03 2002
State-Changed-Why:
    The source code example violates ISO C aliasing rules due to cast.  Either compile with -fno-strict-aliasing or correct the code as described in GCC documentation (e.g., union).

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7632

Re: optimization/7632: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments

[Daniel Jacobowitz]

The following reply was made to PR optimization/7632; it has been noted by GNATS.

From: Daniel Jacobowitz <drow@mvista.com>
To: Horst.Lehser@hightec-rt.com
Cc: gcc-gnats@gcc.gnu.org
Subject: Re: optimization/7632: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments
Date: Mon, 19 Aug 2002 09:33:59 -0400

 On Mon, Aug 19, 2002 at 11:49:16AM -0000, Horst.Lehser@hightec-rt.com wrote:
 > 
 > >Number:         7632
 > >Category:       optimization
 > >Synopsis:       gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments
 > >Confidential:   no
 > >Severity:       critical
 > >Priority:       medium
 > >Responsible:    unassigned
 > >State:          open
 > >Class:          sw-bug
 > >Submitter-Id:   net
 > >Arrival-Date:   Mon Aug 19 04:56:01 PDT 2002
 > >Closed-Date:
 > >Last-Modified:
 > >Originator:     Horst Lehser
 > >Release:        unknown-1.0
 > >Organization:
 > >Environment:
 > 
 > >Description:
 > the following C source will generate incorrect code
 > 
 > double a;
 > double foo(void)
 > {
 > 	((int *)&a)[1] += 5;
 > 	return;
 > }
 > Generated assembler
 > foo:
 >         lis %r3,a+4@ha
 >         lis %r5,a@ha
 >         lwz %r4,a+4@l(%r3)
 >         lfd %f1,a@l(%r5)	/* a not modified yet */
 >         addi %r0,%r4,10
 >         stw %r0,a+4@l(%r3)
 >         blr
 
 (I assume you meant "return a" above).
 
 You may want to read the documentation for -fno-strict-aliasing; your
 code is illegal C.  With -fno-strict-aliasing I get the expected
 result.
 
 -- 
 Daniel Jacobowitz
 MontaVista Software                         Debian GNU/Linux Developer

Re: optimization/7632: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments

[Andrew Pinski]

The following reply was made to PR optimization/7632; it has been noted by GNATS.

From: Andrew Pinski <pinskia@physics.uc.edu>
To: Horst.Lehser@hightec-rt.com
Cc: gcc-gnats@gcc.gnu.org
Subject: Re: optimization/7632: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments
Date: Mon, 19 Aug 2002 08:22:57 -0400

 It is an invalid use of pointers conversion of ISO C aka aliasing rules.
 
 Thanks,
 Andrew Pinski
 
 On Monday, August 19, 2002, at 07:49 , Horst.Lehser@hightec-
 rt.com wrote:
 
 > double a;
 > double foo(void)
 > {
 > 	((int *)&a)[1] += 5;
 > 	return;
 > }
 

optimization/7632: gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments

[Horst.Lehser]

>Number:         7632
>Category:       optimization
>Synopsis:       gcc 3.2 for PowerPC: instruction scheduling ignores dependency in variable asignments
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon Aug 19 04:56:01 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     Horst Lehser
>Release:        unknown-1.0
>Organization:
>Environment:

>Description:
the following C source will generate incorrect code

double a;
double foo(void)
{
	((int *)&a)[1] += 5;
	return;
}
Generated assembler
foo:
        lis %r3,a+4@ha
        lis %r5,a@ha
        lwz %r4,a+4@l(%r3)
        lfd %f1,a@l(%r5)	/* a not modified yet */
        addi %r0,%r4,10
        stw %r0,a+4@l(%r3)
        blr
	

>How-To-Repeat:
compile the following code with
gcc -O3 -S

double a;
double foo(void)
{
	((int *)&a)[1] += 10;
	return a;
}
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted: