Re: G++ optimization bug?

[Larry Smith]

valmosk@gmail.com wrote:
> I posted this in the gcc help forums, but I think it belongs in this
> one since its a g++ issue.
> 
> Hi,
>    I've encountered a problem that may be a compiler optimization bug
> (with branch prediction). So far, I can reproduce it on Linux64 with g
> +
> + 3.4.3:
> 
> nagara:/home/vmoskovi>g++ --version
> g++ (GCC) 3.4.3 20050227 (Red Hat 3.4.3-22.1)
> Copyright (C) 2004 Free Software Foundation, Inc.
> 
> 
> Here is a simple program:
> 
> 
> #define COPYLONG(pDstAddr, pSrcAddr) \
>         ((int *) (pDstAddr))[0] = ((int *)(pSrcAddr))[0]; \
>         ((int *) (pDstAddr))[1] = ((int *)(pSrcAddr))[1];
> 
> 
> #include <iostream.h>
> #include <stdio.h>
> 
> 
> inline int test()
> {
>     long long dataAddr = 5;
>     long long dupAddr = 0;
>     COPYLONG(&dupAddr, &dataAddr);
>     if (dupAddr == 0)
>     {
>         printf("IF\n");
>     }
>     else
>     {
>         printf("ELSE\n");
>     }
> 
> 
>    return 0;
> 
> 
> 
> }
> 
> 
> main()
> {
> 
>   test();
>   return 0;
> 
> 
> 
> }
> 
> 
> The expected output is ELSE since dupAddr  should be set to 5. Here
> is
> the output:
> 
> nagara:/home/vmoskovi>g++ -O1 jt.cpp
> In file included from /usr/lib/gcc/x86_64-redhat-linux/
> 3.4.3/../../../../include/c++/3.4.3/backward/iostream.h:31,
>                  from jt.cpp:7:
> /usr/lib/gcc/x86_64-redhat-linux/3.4.3/../../../../include/c++/3.4.3/
> backward/backward_warning.h:32:2: warning: #warning This file
> includes
> at least one deprecated or antiquated header. Please consider using
> one of the 32 headers found in section 17.4.1.2 of the C++ standard.
> Examples include substituting the <X> header for the <X.h> header for
> C
> ++ includes, or <iostream> instead of the deprecated header
> <iostream.h>. To disable this warning use -Wno-deprecated.
> nagara:/home/vmoskovi>a.out
> ELSE
> nagara:/home/vmoskovi>g++ -O2 jt.cpp
> In file included from /usr/lib/gcc/x86_64-redhat-linux/
> 3.4.3/../../../../include/c++/3.4.3/backward/iostream.h:31,
>                  from jt.cpp:7:
> /usr/lib/gcc/x86_64-redhat-linux/3.4.3/../../../../include/c++/3.4.3/
> backward/backward_warning.h:32:2: warning: #warning This file
> includes
> at least one deprecated or antiquated header. Please consider using
> one of the 32 headers found in section 17.4.1.2 of the C++ standard.
> Examples include substituting the <X> header for the <X.h> header for
> C
> ++ includes, or <iostream> instead of the deprecated header
> <iostream.h>. To disable this warning use -Wno-deprecated.
> nagara:/home/vmoskovi>a.out
> IF
> 
> 
> Using optimization level 01, the output is correct. However, using
> optimization level 02 it takes the wrong branch. I suspect that the
> optimizer doesn't realize that COPYLONG modifies dupAddr. Is this a
> known issue? Has anyone else encountered this problem?
> 

The following corrected code produces "ELSE" with -O1 and -O2.
I don't have the old Redhat platform to test against, nor do I
the old compiler.  Using g++ v4.1.2 with SUSE 10.2 on x86 and
x86_64 the result is always "ELSE".

#include <iostream>
#include <stdio.h>

#define COPYLONG(pDstAddr, pSrcAddr) \
        ((int *) (pDstAddr))[0] = ((int *)(pSrcAddr))[0]; \
        ((int *) (pDstAddr))[1] = ((int *)(pSrcAddr))[1];

inline int test()
{
    // hmmm, pointers on 64 bit hdw are 'long',
    // not 'long long'.
    // int is 32 bits, long is 64 bits, ptr is
    // 64 bits.
    long long dataAddr = 5;
    long long dupAddr = 0;
    COPYLONG(&dupAddr, &dataAddr);
    if (dupAddr == 0)
    {
        printf("IF\n");
    }
    else
    {
        printf("ELSE\n");
    }

   return 0;
}

main()
{
  // print the size of some key data types...
  printf("%u = sizeof int\n", sizeof(int) );
  printf("%u = sizeof long\n", sizeof(long) );
  printf("%u = sizeof ptr\n", sizeof(void *) );
  printf("%u = sizeof long long\n", sizeof(long long) );

  printf("result of test() = ");

  test();

  return 0;
}