[Bug c/21607] New: Optimizations convert 0 to -0 leading to difference in answers.

[Bug c/21607] New: Optimizations convert 0 to -0 leading to difference in answers.

[fsmith at mathworks dot com]

Hi,

Sorry for the missing information above.  gcc -v did not supply the above 
information and for reasons unknown to me gccbug was not installed on the 
system in question.

Compiling with -O changes the results of this program.

double u_im = 0;

extern double atan(double);
extern int printf(const char *str,...);

int main(void)
{
    double y;
    double b;

    /* Becomes -0 with optimizations and 0 without. */
    b = 0 - 3*u_im;
    
    y = atan(-7/b);

    printf("b = %g\n",b);
    printf("atan(-7/b) = %g\n",y);
    
    if(y >= 0) {
        printf("Error happened!\n");
    }
    
    return (y>=0);
}

-- 
           Summary: Optimizations convert 0 to -0 leading to difference in
                    answers.
           Product: gcc
           Version: 3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: fsmith at mathworks dot com
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21607

[Bug c/21607] Optimizations convert 0 to -0 leading to difference in answers.

[fsmith at mathworks dot com]

------- Additional Comments From fsmith at mathworks dot com  2005-05-16 13:48 -------
Created an attachment (id=8901)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8901&action=view)
Minimal self-contained C program exhibiting problem

The answer provided by this program changes depending on whether or
not optimizations are turned on.

The reason for the difference is that the value b changes between 0
and -0 depending on the optimization level.

Here is some sample output:
% gcc emtst_asinh_standalone.c
b = 0
atan(-7/b) = -1.5708
% gcc -O emtst_asinh_standalone.c
b = -0
atan(-7/b) = 1.5708
Error happened!


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21607

[Bug c/21607] Optimizations convert 0 to -0 leading to difference in answers.

[fsmith at mathworks dot com]

------- Additional Comments From fsmith at mathworks dot com  2005-05-16 13:49 -------
Created an attachment (id=8902)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8902&action=view)
Pre-processed version of previous attachment.

Preprocessed version of previous file.	The previous C file was self-contained
so the only benefit you got was deleting my comments.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21607

[Bug c/21607] Optimizations convert 0 to -0 leading to difference in answers.

[fsmith at mathworks dot com]

------- Additional Comments From fsmith at mathworks dot com  2005-05-16 13:50 -------
Information from gcc -save-temps -v follows:

Reading specs from /usr/libexec/gcc/darwin/ppc/3.3/specs
Thread model: posix
gcc version 3.3 20030304 (Apple Computer, Inc. build 1495)
 /usr/libexec/gcc/darwin/ppc/3.3/cc1 -E -quiet -v -D__GNUC__=3 -
D__GNUC_MINOR__=3 -D__GNUC_PATCHLEVEL__=0 -D__APPLE_CC__=1495 -D__DYNAMIC__ 
emtst_asinh_standalone.c -fPIC -O emtst_asinh_standalone.i
ignoring nonexistent directory "/usr/ppc-darwin/include"
ignoring nonexistent directory "/Local/Library/Frameworks"
#include "..." search starts here:
#include <...> search starts here:
 /usr/local/include
 /usr/include/gcc/darwin/3.3
 /usr/include
End of search list.
Framework search starts here:
 /System/Library/Frameworks
 /Library/Frameworks
End of framework search list.
 /usr/libexec/gcc/darwin/ppc/3.3/cc1 -fpreprocessed emtst_asinh_standalone.i -
fPIC -quiet -dumpbase emtst_asinh_standalone.c -auxbase 
emtst_asinh_standalone -O -version -o emtst_asinh_standalone.s
GNU C version 3.3 20030304 (Apple Computer, Inc. build 1495) (ppc-darwin)
	compiled by GNU C version 3.3 20030304 (Apple Computer, Inc. build 
1495).
GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=131072
 /usr/libexec/gcc/darwin/ppc/as -arch ppc -o emtst_asinh_standalone.o 
emtst_asinh_standalone.s
 ld -arch ppc -dynamic -o a.out -lcrt1.o -lcrt2.o -L/usr/lib/gcc/darwin/3.3 -
L/usr/lib/gcc/darwin -L/usr/libexec/gcc/darwin/ppc/3.3/../../.. 
emtst_asinh_standalone.o -lgcc -lSystem |
 c++filt3


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
  GCC build triplet|                            |1495


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21607

[Bug c/21607] Optimizations convert 0 to -0 leading to difference in answers.

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-05-16 13:53 -------
This is a bug in Apple's GCC, it uses fmadd instructions when it is not safe, please report this to Apple, 
I think it is fixed in Apple's 4.0.0 GCC based compiler.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21607

[Bug c/21607] Optimizations convert 0 to -0 leading to difference in answers.

[fsmith at mathworks dot com]

------- Additional Comments From fsmith at mathworks dot com  2005-05-16 14:04 -------
How do I report it to Apple?  On their site

http://developer.apple.com/documentation/DeveloperTools/gcc-3.3/gcc/gccbug.html#gccbug

they told me to report it here (effectively).  But since, for reasons unknown to
me, gccbug wasn't installed on the MAC I was using I went straight to bugzilla.

-Fred

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21607

[Bug c/21607] Optimizations convert 0 to -0 leading to difference in answers.

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-05-16 14:08 -------
(In reply to comment #5)
> How do I report it to Apple?  On their site
> 
> http://developer.apple.com/documentation/DeveloperTools/gcc-3.3/gcc/gccbug.html#gccbug

http://bugreport.apple.com

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21607