This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

optimization/4751: g++ -O producing incorrect code for ((global1&255)==255 || global2)



>Number:         4751
>Category:       optimization
>Synopsis:       g++ -O producing incorrect code for ((global1&255)==255 || global2)
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Tue Oct 30 15:46:00 PST 2001
>Closed-Date:
>Last-Modified:
>Originator:     keenan@ghs.com
>Release:        gcc version 3.0.1 (also reproduced on 2.95.2 and 2.95.3)
>Organization:
>Environment:
HP-UX 11.0 (also reproduced on 10.20)
>Description:
An illegal optimization is being performed, such that the attached file produces incorrect output when compiled with -O (or -O1, -O2, -O3). The simple test case breaks down to code that is of the form (where both globals are type int)

if(((global1&255)==global1) || global2)
    exectute_some_statement();

The statement is _always_ executed, even when both expressions are false. If split into two tests:
if(((global1&255)==global1))
    exectute_some_statement();
if(global2)
    exectute_some_statement();

The statment is not executed.

I wasn't able to reproduce this bug on non-HP-UX hosts.
>How-To-Repeat:
g++ -O foo.cc
./a.out
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: text/plain; name="foo.cc"
Content-Disposition: inline; filename="foo.cc"

#include <stdio.h>
int global_0 = 0;
int global_neg1 = -1;

int main()
{
    bool test = ((global_neg1&255)==global_neg1);
//    bool test = !(global_neg1&~255);
    if(test || global_0) {
	printf ("Guess again...\n");
    }
    return 0;
}


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]