[Bug other/23469] New: Behaviour of built-in __signbitf(x) differs with optimization

[Bug other/23469] New: Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

The return value of the builtin signbit(x) macro (called with a float argument)
differs when compiling with optimizations or not

When optimizations are off the return value is 0x8000 or 0 (signbit set and
not-set, respectively).

When optimizations are on (-O1) return value is 1 or 0 (signbit set and not-set,
respectively).

 I'll attach a sample .cpp file, .ii file, and the output of -save-temps to this
report.

 The sample .cpp file shows a simple example. The function mysignbit() is
defined just as the __signbitf(x) function in include/bits/mathinline.h but
returns a different value than __signbitf(x) when compiling without
optimizations, but returns the same value when compiling with optimizations (-O1
and above).

-- 
           Summary: Behaviour of built-in __signbitf(x) differs with
                    optimization
           Product: gcc
           Version: 3.4.4
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: other
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: ddneilson at gmail dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-18 20:43 -------
Created an attachment (id=9535)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9535&action=view)
Source file of example


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-18 20:44 -------
Created an attachment (id=9536)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9536&action=view)
.ii file generated by compiling with: g++ -Wall -v -save-temps signbit.cpp


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-18 20:45 -------
Created an attachment (id=9537)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9537&action=view)
Ouput from compile with: g++ -Wall -v -save-temps signbit.cpp


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-18 20:46 -------
Created an attachment (id=9538)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9538&action=view)
.ii file generated by compiling with: g++ -Wall -v -O1 -save-temps signbit.cpp


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-18 20:47 -------
Created an attachment (id=9539)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9539&action=view)
Output from compile with: g++ -Wall -v -O1 -save-temps signbit.cpp


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-18 20:49 -------
Created an attachment (id=9540)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9540&action=view)
Output from running the program without optimizations


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Attachment #9537|Ouput from compile with: g++|Output from compile with:
        description|-Wall -v -save-temps        |g++ -Wall -v -save-temps
                   |signbit.cpp                 |signbit.cpp


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-18 20:49 -------
Created an attachment (id=9541)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9541&action=view)
Output from running the program with -O1


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-08-18 21:41 -------
This is not a bug as signbit just have to return a non zero value if it has the value passed has its sign 
bit set.
So defining it this way:
inline int mysignbit(float x) {
  union { float f; unsigned int i; } u;
  u.f = x;
  return u.i&0x80000000UL;
}

Is also legal.

The reason for the different is optimizing for size vs optimizing for speed.  The definition you gave is 
faster but larger.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[ddneilson at gmail dot com]

------- Additional Comments From ddneilson at gmail dot com  2005-08-21 20:28 -------
One of the really odd things, that makes me still think this is a bug in some
way is that the definition of "mysignbit()" that I gave in signbit.cpp is
identical to the definition of __signbitf(x) as given in bits/mathinline.h. But,
yet, when optimizations aren't turned on the two functions have differing return
values; mysignbit() will return 1 iff the signbit is set, whereas __signbitf()
will return 0x80000000 iff the signbit is set.

 Furthermore, when all documented -O1 flags are turned on manually (without
actually using -O1) the output of __signbitf(x) is the same as the unoptimized
version; it only differs when one of the -On flags are used.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469

[Bug other/23469] Behaviour of built-in __signbitf(x) differs with optimization

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-08-21 20:31 -------
Again this is not a bug as signbitf says about nonzero value and nothing about the exact value.  In fact 
it changes based on optimizations is because well some times it is for size reasons, look at -Os and 
you will see that.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23469