Re: optimization/7041: a.out gives different results when compiled by g++ 3.1 and g++ 3.1 -O

Re: optimization/7041: a.out gives different results when compiled by g++ 3.1 and g++ 3.1 -O

[Andrew Pinski]

The following reply was made to PR optimization/7041; it has been noted by GNATS.

From: Andrew Pinski <pinskia@physics.uc.edu>
To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, gcc-bugs@gcc.gnu.org,
   nobody@gcc.gnu.org, alphonse@lri.fr
Cc:  
Subject: Re: optimization/7041: a.out gives different results when compiled by g++ 3.1 and g++ 3.1 -O
Date: Sun, 6 Oct 2002 21:54:27 -0700

 Looks fixed for 3.3 (gcc version 3.3 20021003 (experimental) ):
 hodgepodge:~/src/gnu/gcctest>!g+
 g++ pr7041.cc
 hodgepodge:~/src/gnu/gcctest>g++ pr7041.cc -Wall
 pr7041.cc: In function `int main()':
 pr7041.cc:63: warning: unused variable `prng gen'
 hodgepodge:~/src/gnu/gcctest>./a.out
 2629073562d
 hodgepodge:~/src/gnu/gcctest>g++ pr7041.cc -Wall -O
 pr7041.cc: In function `int main()':
 pr7041.cc:63: warning: unused variable `prng gen'
 ^[[Ahodgepodge:~/src/gnu/gcctest>./a.out
 2983301384d
 hodgepodge:~/src/gnu/gcctest>
 hodgepodge:~/src/gnu/gcctest>g++ pr7041.cc -Wall -O -march=i586
 pr7041.cc: In function `int main()':
 pr7041.cc:63: warning: unused variable `prng gen'
 ^[[Ahodgepodge:~/src/gnu/gcctest>./a.out
 2983301384d
 hodgepodge:~/src/gnu/gcctest>g++ pr7041.cc -Wall -O -march=i386
 pr7041.cc: In function `int main()':
 pr7041.cc:63: warning: unused variable `prng gen'
 hodgepodge:~/src/gnu/gcctest>!./
 ./a.out
 2983301384d
 hodgepodge:~/src/gnu/gcctest>
 
 
 
 
 http://gcc.gnu.org/cgi-bin/gnatsweb.pl
 

Re: optimization/7041: a.out gives different results when compiled by g++ 3.1 and g++ 3.1 -O

[rth]

Synopsis: a.out gives different results when compiled by g++ 3.1 and g++ 3.1 -O

State-Changed-From-To: open->closed
State-Changed-By: rth
State-Changed-When: Mon Oct  7 00:16:06 2002
State-Changed-Why:
    Fixed.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7041

optimization/7041: a.out gives different results when compiled by g++ 3.1 and g++ 3.1 -O

[alphonse]

>Number:         7041
>Category:       optimization
>Synopsis:       a.out gives different results when compiled by g++ 3.1 and g++ 3.1 -O
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Jun 15 10:46:00 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     alphonse
>Release:        g++-3.1.1 (GCC) 3.1.1 (Mandrake Linux 8.3 3.1.1-0.2mdk)
>Organization:
>Environment:
Reading specs from /usr/lib/gcc-lib/i586-mandrake-linux-gnu/3.1.1/specs
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --enable-shared --enable-threads=posix --disable-checking --enable-long-long --with-gxx-include-dir=/usr/include/g++-v31 --enable-languages=c,c++,f77,objc,java --host=i586-mandrake-linux-gnu --with-system-zlib
Thread model: posix
gcc version 3.1.1 (Mandrake Linux 8.3 3.1.1-0.2mdk)
>Description:
the behaviour of a C++ version of the mersenne-twister pseudo-random number generator is wrong when compiled with -O (and greater). The C version works fine.

I have simplified the source code in such a way it outputs only the first number.
I reproduce the problem by running :
1066# g++ -Wall test.cpp
test.cpp: In function `int main()':
test.cpp:63: warning: unused variable `prng gen'
1067# a.out
2629073562d    /* correct result */
1068# g++ -Wall -O test.cpp
test.cpp: In function `int main()':
test.cpp:63: warning: unused variable `prng gen'
1069# a.out
2983301384d    /* incorrect result */

Congratulation for your work,
Erick Alphonse.
>How-To-Repeat:
#include <cstdio>

typedef unsigned long ui32;

class prng {
  enum {N = 624,              // length of state vector
	M = 397,              // period parameter
	MAGIC = 0x9908B0DFU}; // magic constant

  ui32 state[N];              // internal state
  ui32 *next;                 // next value to get from state
  int left;                   // number of values left before reload needed

  ui32 hiBit(const ui32 u) {return u&0x80000000U;}  // mask all but highest bit of u
  ui32 loBit(const ui32 u) {return u&0x00000001U;}  // mask all but lowest bit of u
  ui32 loBits(const ui32 u) {return u&0x7fffffffU;} // mask the highest bit of u 

  ui32 mixBits(const ui32 u,const ui32 v) {return hiBit(u) | loBits(v);}
  ui32 twist(const ui32 m,const ui32 s0,const ui32 s1) {
    return m ^ (mixBits(s0,s1)>>1) ^ (loBit(s1) ? MAGIC : 0U);
  }

  // Generate N new values in state
  void reload();

public:
  prng(const ui32 seed); // initializes with a simple ui32
};

inline void prng::reload() {
  register ui32 *p = state;
  register int i;

  for( i = N - M; i--; )
    *p++ = twist( p[M], p[0], p[1] );
  for( i = M; --i; )
    *p++ = twist( p[M-N], p[0], p[1] );
  *p = twist( p[M-N], p[0], state[0] );

  left = N, next = state;

  printf("%lud\n",*state);
}

inline prng::prng(const ui32 seed) {
  register ui32 j;

  state[0]= seed & 0xffffffffUL;
  for (j=1; j<N; j++) {
    state[j] = (1812433253UL * (state[j-1] ^ (state[j-1] >> 30)) + j); 
    /* See Knuth TAOCP Vol2. 3rd Ed. P.106 for multiplier. */
    /* In the previous versions, MSBs of the seed affect   */
    /* only MSBs of the array state[].                        */
    /* 2002/01/09 modified by Makoto Matsumoto             */
    state[j] &= 0xffffffffUL;  /* for >32 bit machines */
  }

  reload();
}

int main()
{
    prng gen(1);
}
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: application/octet-stream; name="test.ii.gz"
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename="test.ii.gz"
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