[Bug optimization/9196] [3.4 regression] Inlining of function with void * arithmetic is misoptimized

[Bug optimization/9196] [3.4 regression] Inlining of function with void * arithmetic is misoptimized

[pinskia@physics.uc.edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196


pinskia@physics.uc.edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2


------- Additional Comments From pinskia@physics.uc.edu  2003-06-03 19:14 -------
Moving up priority because it effects glibc.



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug optimization/9196] [3.4 regression] Inlining of function with void * arithmetic is misoptimized

[aj@gcc.gnu.org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196



------- Additional Comments From aj@gcc.gnu.org  2003-06-09 06:02 -------
Bug confirmed with gcc (GCC) 3.4 20030604 (experimental).

[Bug optimization/9196] [3.4 regression] Inlining of function with void * arithmetic is misoptimized

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|0000-00-00 00:00:00         |2003-07-05 03:07:33
               date|                            |


------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-05 03:07 -------
still happens on the mainline (20030704).

[Bug optimization/9196] [3.4 regression] Inlining of function with void * arithmetic is misoptimized

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
           Priority|P2                          |P1
   Last reconfirmed|2003-07-05 03:07:33         |2003-07-28 01:17:21
               date|                            |


------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-28 01:17 -------
Still happens on the mainline (20030727).

[Bug optimization/9196] [3.4 regression] Inlining of function with void * arithmetic is misoptimized

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2003-07-28 01:17:21         |2003-08-11 02:55:06
               date|                            |


------- Additional Comments From pinskia at gcc dot gnu dot org  2003-08-11 02:55 -------
It is not the char * arithmetic that is being miscompiled as if I replace the void* inside __memrchr 
with char*, I still get the miscompile.  Also the code is wrong it uses D when it really meant to use 
d and it should be using "+" instead of "=" aqnd it has enouns additions of accesing the memory.
__inline void *
__memrchr (__const char *__s, int __c, size_t __n)
{
  register unsigned long int __d0;

  register char *__res;
  if (__n == 0)
    return ((void *)0);
  __res = __s + __n - 1;
  __d0 = __n;

  __asm__ __volatile__
    ("std\n\t"
     "repne; scasb\n\t"
     "je 1f\n\t"
     "orl $-1,%0\n"
     "1:\tcld"
     : "+d" (__res), "+c" (__d0)
     : "a" (__c)
     : "cc");

  return (void*)(__res + 1);
}

If I change "register char *__res;" to "register int __res;", it works so it looks like it is the pointer 
which causes this.

Also it goes wrong in combine (GCC sets __res dead after asm for some reason).

[Bug optimization/9196] [3.4 regression] Inlining of function with pointer arithmetic is misoptimized

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[3.4 regression] Inlining of|[3.4 regression] Inlining of
                   |function with void *        |function with  pointer
                   |arithmetic is misoptimized  |arithmetic is misoptimized


------- Additional Comments From pinskia at gcc dot gnu dot org  2003-08-11 18:02 -------
Change summary based on my anlysis last night.

[Bug optimization/9196] [3.4 regression] pointer arithmetic is misoptimized

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2003-08-11 02:55:06         |2003-08-23 18:57:02
               date|                            |
            Summary|[3.4 regression] Inlining of|[3.4 regression] pointer
                   |function with  pointer      |arithmetic is misoptimized
                   |arithmetic is misoptimized  |


------- Additional Comments From pinskia at gcc dot gnu dot org  2003-08-23 18:57 -------
The inner has nothing to do with this bug as I can reproduce it when I inlined it myself:
static void test_memrchr (void)
{
  it = "memrchr";
  void *ret;
  {
  __const char *__s = "abcd";
  int __c = 'z';
  size_t __n = 5;
{
  register unsigned long int __d0;

  register void *__res;
  if (__n == 0)
    { 
      ret = (void*)0;
      goto end;
    }
  __res = __s + __n - 1;
  __d0 = __n;

  __asm__ __volatile__
    ("std\n\t"
     "repne; scasb\n\t"
     "je 1f\n\t"
     "orl $-1,%0\n"
     "1:\tcld"
     : "+d" (__res), "+c" (__d0)
     : "a" (__c)
     : "cc");

  ret = (void*)(__res + 1);
}
  goto end;
  }
end:
  check (ret  == ((void *)0), 1);
}

[Bug optimization/9196] [3.4 regression] pointer arithmetic is misoptimized

[zack at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9196


zack at gcc dot gnu dot org changed:

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


------- Additional Comments From zack at gcc dot gnu dot org  2003-08-25 20:10 -------
The C semantics for pointer arithmetic mean that GCC is allowed to assume that
this line

  return __res + 1;

never causes a null pointer to be returned.  The "mis-optimized" code has been
optimized based on that assumption.  In short, the bug is in glibc.

The simplest fix is to write "incl %0" as the last instruction of the assembly
block rather than trying to do this calculation in C.