optimization/4867: gcc-3.1 20011112 generates wrong code for integer division

[pavenis at latnet dot lv]

>Number:         4867
>Category:       optimization
>Synopsis:       gcc-3.1 20011112 generates wrong code for integer division
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Tue Nov 13 09:26:01 PST 2001
>Closed-Date:
>Last-Modified:
>Originator:     Andris Pavenis
>Release:        gcc-3.1 20011112 (experimental), gcc-3.1 20011030 (experimental)
>Organization:
>Environment:
gcc-3.1 20011112 for i686-pc-linux-gnu,
gcc-3.1 20011030 for i586-pc-msdosdjgpp
>Description:
gcc-3.1 generates bad code for integer division and taking
remainder when result is being stored in structure members
which size is specified in bits (8 bits in this case) in
included example. It happens for -march=i386, -march=i486,
-march=i586 and -march=k6, but not for -march=i686.

I'm getting output:
245 19

instead of correct:
42 1

Andris

#include <stdio.h>

struct foo { unsigned a:8, b:8; };

struct foo div10 (unsigned short arg);

int main (void)
{
    struct foo bar = div10(421);
    printf ("%u %u\n", (unsigned) bar.a, (unsigned) bar.b);
}

struct foo div10 (unsigned short arg)
{
    struct foo bar;
    bar.a = arg/10;
    bar.b = arg%10;
    return bar;
}




>How-To-Repeat:
gcc -O2 -march=i386 foo.c -o foo
./foo

(the same for i486, i586 and k6).
>Fix:

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