Re: optimization/7175: Calculation yields different results for version optimized with -O2

Re: optimization/7175: Calculation yields different results for version optimized with -O2

[bangerth]

Synopsis: Calculation yields different results for version optimized with -O2

State-Changed-From-To: open->feedback
State-Changed-By: bangerth
State-Changed-When: Fri Dec  6 13:50:40 2002
State-Changed-Why:
    This code works for me with gcc3.2. Would you mind checking
    whether this helps on your system as well?
    
    Thanks
      Wolfgang

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7175

Re: optimization/7175: Calculation yields different results for version optimized with -O2

[bangerth]

Synopsis: Calculation yields different results for version optimized with -O2

State-Changed-From-To: feedback->closed
State-Changed-By: bangerth
State-Changed-When: Tue Dec 10 06:28:23 2002
State-Changed-Why:
    Then this is not a problem of handling inaccuracies, but of
    handling excess accuracy due to the idiotic x87 floating
    point representation. Read the section on non-bugs on the
    gcc homepage for a description.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7175

Re: optimization/7175: Calculation yields different results for version optimized with -O2

[Gunther H Weber]

The following reply was made to PR optimization/7175; it has been noted by GNATS.

From: Gunther H Weber <weber@informatik.uni-kl.de>
To: bangerth@dealii.org, gcc-bugs@gcc.gnu.org, gcc-prs@gcc.gnu.org,
   nobody@gcc.gnu.org, weber@informatik.uni-kl.de, gcc-gnats@gcc.gnu.org
Cc:  
Subject: Re: optimization/7175: Calculation yields different results for version
 optimized with -O2
Date: Mon, 09 Dec 2002 20:13:39 -0800

 I still get the problem with gcc3.2. However, it disappears as soon as I 
 use -ffloat-store. So it seems to be due to poor handling of numerical 
 inaccuracies.
 
 Thank you for your feedback
    Gunther
 
 bangerth@dealii.org wrote:
 
 >Synopsis: Calculation yields different results for version optimized with -O2
 >
 >State-Changed-From-To: open->feedback
 >State-Changed-By: bangerth
 >State-Changed-When: Fri Dec  6 13:50:40 2002
 >State-Changed-Why:
 >    This code works for me with gcc3.2. Would you mind checking
 >    whether this helps on your system as well?
 >    
 >    Thanks
 >      Wolfgang
 >
 >http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7175
 >  
 >
 
 
 

Re: optimization/7175: Calculation yields different results for version optimized with -O2

[Michael Veksler]

The following reply was made to PR optimization/7175; it has been noted by GNATS.

From: Michael Veksler <veksler@il.ibm.com>
To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, gcc-bugs@gcc.gnu.org,
   nobody@gcc.gnu.org, weber@informatik.uni-kl.de
Cc:  
Subject: Re: optimization/7175: Calculation yields different results for version
 optimized with -O2
Date: Sat, 10 Aug 2002 06:29:05 +0300

 http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7175
 
 The moment you will add at the beginning of your main():
   fpu_control_t fpu_bits;
   _FPU_GETCW(fpu_bits);
   fpu_bits &= ~_FPU_EXTENDED;
   fpu_bits |= _FPU_DOUBLE;
   _FPU_SETCW(fpu_bits);
 
 Assuming you init "res" to zeros, you are goint to get a consistent
 result.
 This probably means that this PR is basically the same as PR/6914.
 

optimization/7175: Calculation yields different results for version optimized with -O2

[weber]

>Number:         7175
>Category:       optimization
>Synopsis:       Calculation yields different results for version optimized with -O2
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Mon Jul 01 04:56:01 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     Gunther Weber
>Release:        gcc version 3.1
>Organization:
>Environment:
RedHat Linux 7.3 running on a 800 MHz Celeron (Coppermine) processor. Problem also occurs on an AMD Athlon(tm) XP 1700+ system.
GCC Build option:
Reading specs from /usr/global/lib/gcc-lib/i586-pc-linux-gnu/3.1/specs
Configured with: ./configure --prefix=/usr/global --enable-threads=posix
Thread model: posix
gcc version 3.1
>Description:
A program that calculates an expression yields different results when compiled with -O2 option as compared to -O1 or no optimization. The file attachment contains code with everything but the necessary code to reproduce the bug was removed. Note that trying to print variable conntents for debugging purposes gets rid of the bug. A complete log of compiling the program and its results is listed under How-To-Reperat.
>How-To-Repeat:
[weber@stella TestGCC]$ g++3.1 test.cc
[weber@stella TestGCC]$ ./a.out
Using second solution t=0
0.131997, 0.69765
[weber@stella TestGCC]$ g++3.1 -O1 test.cc
[weber@stella TestGCC]$ ./a.out
Using second solution t=0
0.131997, 0.69765
[weber@stella TestGCC]$ g++3.1 -O2 test.cc
[weber@stella TestGCC]$ ./a.out
Using second solution t=4.14165e-17
0.568462, -0.302107
[weber@stella TestGCC]$ g++3.1 -O2 -DPRINT_T test.cc
[weber@stella TestGCC]$ ./a.out
t1=2 t2=0
Using second solution t=0
0.131997, 0.69765
>Fix:
None that I know of except not to use -O2 option.
>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: application/octet-stream; name="test.cc"
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename="test.cc"
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