[Bug c/44683] New: Optimization bug with copysign builtin

[Bug c/44683] New: Optimization bug with copysign builtin

[dickinsm at gmail dot com]

Here's a minimal test-case:

---------


int copysign_bug(double x)
{
  if (x * 0.5 == x)
    return 1;
  else if (copysign(1.0, x) < 0.0)
    return 2;
  else
    return 3;
}

int main(void) {
  double x;
  x = -0.0;
  printf("copysign_bug(%.17g) = %d\n", x, copysign_bug(x));

  x = 0.0;
  printf("copysign_bug(%.17g) = %d\n", x, copysign_bug(x));

  return 0;
}

-------

I'm compiling and executing the above code (filename copysign_bug.c) on
OpenSolaris build snv_134 running in Parallels on a Macbook Pro (Intel Core 2
Duo), using the following command line:

gcc-4.4 -m64 -fno-inline -g -O3 copysign_bug.c -o copysign_bug &&
./copysign_bug

I get the results:

copysign_bug(-0) = 3
copysign_bug(0) = 3

I was expecting to see the results:

copysign_bug(-0) = 2
copysign_bug(0) = 3

I get the latter results if I omit the '-O3' flag, or if I compile with
-fno-builtin-copysign and link with the math library.

gcc build information:

dickinsm@eratosthenes:~/trunk$ gcc-4.4 -v -save-temps -m64 -lm
-fno-builtin-copysign -fno-inline -g copysign_bug.c
Using built-in specs.
Target: i386-pc-solaris2.11
Configured with: ../gcc-4.4.4/configure --prefix=/usr/local
--program-suffix=-4.4 --with-mpfr-include=/usr/include/mpfr
--with-gmp-include=/usr/include/gmp --with-as=/usr/bin/gas --with-gnu-as
--with-ld=/usr/bin/ld --without-gnu-ld --enable-shared --enable-multilib
--enable-languages=c,c++,objc
Thread model: posix
gcc version 4.4.4 (GCC) 
COLLECT_GCC_OPTIONS='-v' '-save-temps' '-m64' '-fno-builtin-copysign'
'-fno-inline' '-g' '-mtune=generic'
 /usr/local/libexec/gcc/i386-pc-solaris2.11/4.4.4/cc1 -E -quiet -v -imultilib
amd64 copysign_bug.c -m64 -mtune=generic -fno-builtin-copysign -fno-inline -g
-fworking-directory -fpch-preprocess -o copysign_bug.i
ignoring nonexistent directory
"/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/../../../../i386-pc-solaris2.11/include"
#include "..." search starts here:
#include <...> search starts here:
 /usr/local/include
 /usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/include
 /usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/include-fixed
 /usr/include
End of search list.
COLLECT_GCC_OPTIONS='-v' '-save-temps' '-m64' '-fno-builtin-copysign'
'-fno-inline' '-g' '-mtune=generic'
 /usr/local/libexec/gcc/i386-pc-solaris2.11/4.4.4/cc1 -fpreprocessed
copysign_bug.i -quiet -dumpbase copysign_bug.c -m64 -mtune=generic -auxbase
copysign_bug -g -version -fno-builtin-copysign -fno-inline -o copysign_bug.s
GNU C (GCC) version 4.4.4 (i386-pc-solaris2.11)
        compiled by GNU C version 4.4.4, GMP version 4.3.1, MPFR version
2.4.1-p5.
GGC heuristics: --param ggc-min-expand=99 --param ggc-min-heapsize=131015
Compiler executable checksum: f408a824c22e56e60b321448197fb865
COLLECT_GCC_OPTIONS='-v' '-save-temps' '-m64' '-fno-builtin-copysign'
'-fno-inline' '-g' '-mtune=generic'
 /usr/bin/gas -v -V -Qy --64 -s -o copysign_bug.o copysign_bug.s
GNU assembler version 2.19 (i386-pc-solaris2.11) using BFD version (GNU
Binutils) 2.19
COMPILER_PATH=/usr/local/libexec/gcc/i386-pc-solaris2.11/4.4.4/:/usr/local/libexec/gcc/i386-pc-solaris2.11/4.4.4/:/usr/local/libexec/gcc/i386-pc-solaris2.11/:/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/:/usr/local/lib/gcc/i386-pc-solaris2.11/:/usr/ccs/bin/
LIBRARY_PATH=/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/amd64/:/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/../../../amd64/:/lib/amd64/:/usr/lib/amd64/:/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/:/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/../../../:/lib/:/usr/lib/
COLLECT_GCC_OPTIONS='-v' '-save-temps' '-m64' '-fno-builtin-copysign'
'-fno-inline' '-g' '-mtune=generic'
 /usr/local/libexec/gcc/i386-pc-solaris2.11/4.4.4/collect2 -V -Y
P,/lib/64:/usr/lib/64 -Qy /usr/lib/amd64/crt1.o /usr/lib/amd64/crti.o
/usr/lib/amd64/values-Xa.o
/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/amd64/crtbegin.o
-L/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/amd64
-L/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/../../../amd64 -L/lib/amd64
-L/usr/lib/amd64 -L/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4
-L/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/../../.. -lm copysign_bug.o
-lgcc -lgcc_eh -lc -lgcc -lgcc_eh
/usr/local/lib/gcc/i386-pc-solaris2.11/4.4.4/amd64/crtend.o
/usr/lib/amd64/crtn.o
ld: Software Generation Utilities - Solaris Link Editors: 5.11-1.1689


-- 
           Summary: Optimization bug with copysign builtin
           Product: gcc
           Version: 4.4.4
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: dickinsm at gmail dot com
 GCC build triplet: i386-pc-solaris2.11
  GCC host triplet: i386-pc-solaris2.11
GCC target triplet: i386-pc-solaris2.11


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug c/44683] Optimization bug with copysign builtin

[dickinsm at gmail dot com]

------- Comment #1 from dickinsm at gmail dot com  2010-06-26 17:01 -------
Created an attachment (id=21012)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21012&action=view)
Preprocessed source file


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug c/44683] Optimization bug with copysign builtin

[dickinsm at gmail dot com]

------- Comment #2 from dickinsm at gmail dot com  2010-06-26 18:16 -------
I'm also seeing this bug in gcc 4.5.0 on OS X, and it's been independently
confirmed in gcc 4.4.3 on Fedora 12 (see http://bugs.python.org/issue9069).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug c/44683] Optimization bug with copysign builtin

[dickinsm at gmail dot com]

------- Comment #3 from dickinsm at gmail dot com  2010-06-26 18:26 -------
Apologies;  I seem to have messed up the original bug report, by accidentally
pasting a slightly edited version of the original bug file.  The minimal
test-case is as follows:


#include <math.h>
#include <stdio.h>

int copysign_bug(double x)
{
  if (x && (x * 0.5 == x))    // note the extra "x &&" here
    return 1;
  if (copysign(1.0, x) < 0.0)
    return 2;
  else
    return 3;
}

int main(void) {
  double x;
  x = -0.0;
  printf("copysign_bug(%.17g) = %d\n", x, copysign_bug(x));

  x = 0.0;
  printf("copysign_bug(%.17g) = %d\n", x, copysign_bug(x));

  return 0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug rtl-optimization/44683] Optimization bug with copysign builtin

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2010-06-26 22:57 -------
Confirmed.  Somewhere during RTL opts we thread a jump the wrong way.  Also
happens with SSE math.

int __attribute__((noinline,noclone))
copysign_bug (double x)
{
  if (x != 0.0 && (x * 0.5 == x))
    return 1;
  if (__builtin_copysign(1.0, x) < 0.0)
    return 2;
  else
    return 3;
}
int main(void)
{
  double x = -0.0;
  if (copysign_bug (x) != 2)
    __builtin_abort ();
  return 0;
}


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
          Component|c                           |rtl-optimization
     Ever Confirmed|0                           |1
 GCC target triplet|i386-pc-solaris2.11         |i?86-*-*
           Keywords|                            |wrong-code
      Known to fail|                            |4.6.0
   Last reconfirmed|0000-00-00 00:00:00         |2010-06-26 22:57:46
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug rtl-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.3.4
            Summary|Optimization bug with       |[4.4/4.5/4.6 Regression]
                   |copysign builtin            |Optimization bug with
                   |                            |copysign builtin
   Target Milestone|---                         |4.4.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug rtl-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[hjl dot tools at gmail dot com]

------- Comment #5 from hjl dot tools at gmail dot com  2010-06-27 03:04 -------
It is caused by revision 134833:

http://gcc.gnu.org/ml/gcc-cvs/2008-04/msg01062.html


-- 

hjl dot tools at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |espindola at google dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug tree-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2010-06-27 08:50 -------
Indeed.  It's exposed by that patch.

DOM, from seeing x_2(D) != 0.0 records x_2(D) == 0.0 as equivalence.  This
is not correct if signed zeros can appear.

I have a patch.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rguenth at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
          Component|rtl-optimization            |tree-optimization
   Last reconfirmed|2010-06-26 22:57:46         |2010-06-27 08:50:38
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug tree-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[rguenth at gcc dot gnu dot org]

------- Comment #7 from rguenth at gcc dot gnu dot org  2010-06-27 11:16 -------
Subject: Bug 44683

Author: rguenth
Date: Sun Jun 27 11:16:46 2010
New Revision: 161467

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=161467
Log:
2010-06-27  Richard Guenther  <rguenther@suse.de>

        PR tree-optimization/44683
        * tree-ssa-dom.c (record_edge_info): Record equivalences for the
        false edge from the inverted condition.

        * gcc.c-torture/execute/pr44683.c: New testcase.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr44683.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-dom.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug tree-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2010-06-27 11:21 -------
Subject: Bug 44683

Author: rguenth
Date: Sun Jun 27 11:21:07 2010
New Revision: 161468

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=161468
Log:
2010-06-27  Richard Guenther  <rguenther@suse.de>

        PR tree-optimization/44683
        * tree-ssa-dom.c (record_edge_info): Record equivalences for the
        false edge from the inverted condition.

        * gcc.c-torture/execute/pr44683.c: New testcase.

Added:
    branches/gcc-4_5-branch/gcc/testsuite/gcc.c-torture/execute/pr44683.c
Modified:
    branches/gcc-4_5-branch/gcc/ChangeLog
    branches/gcc-4_5-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_5-branch/gcc/tree-ssa-dom.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug tree-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[rguenth at gcc dot gnu dot org]

------- Comment #9 from rguenth at gcc dot gnu dot org  2010-06-27 11:22 -------
Subject: Bug 44683

Author: rguenth
Date: Sun Jun 27 11:22:06 2010
New Revision: 161469

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=161469
Log:
2010-06-27  Richard Guenther  <rguenther@suse.de>

        PR tree-optimization/44683
        * tree-ssa-dom.c (record_edge_info): Record equivalences for the
        false edge from the inverted condition.

        * gcc.c-torture/execute/pr44683.c: New testcase.

Added:
    branches/gcc-4_4-branch/gcc/testsuite/gcc.c-torture/execute/pr44683.c
Modified:
    branches/gcc-4_4-branch/gcc/ChangeLog
    branches/gcc-4_4-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_4-branch/gcc/tree-ssa-dom.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug tree-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[rguenth at gcc dot gnu dot org]

------- Comment #10 from rguenth at gcc dot gnu dot org  2010-06-27 11:23 -------
Fixed.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
      Known to fail|4.6.0                       |4.4.4 4.5.0
      Known to work|4.3.4                       |4.3.4 4.4.5 4.5.1 4.6.0
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

[Bug tree-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

--- Comment #12 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-23 15:57:13 UTC ---
Author: ebotcazou
Date: Sun Oct 23 15:57:09 2011
New Revision: 180340

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=180340
Log:
    * fold-const.c (invert_tree_comparison): Always invert EQ_EXPR/NE_EXPR.

    PR tree-optimization/44683
    * tree-ssa-dom.c (record_edge_info): Record simple equivalences only if
    we can be sure that there are no signed zeros involved.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/fold-const.c
    trunk/gcc/tree-ssa-dom.c

[Bug tree-optimization/44683] [4.4/4.5/4.6 Regression] Optimization bug with copysign builtin

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44683

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ebotcazou at gcc dot
                   |                            |gnu.org
      Known to work|                            |
      Known to fail|                            |

--- Comment #11 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-22 16:56:59 UTC ---
While working on a related subject, the testcase started to regress.  IMHO the
fix only papers over the problem (try with -fno-trapping-math for example).
Even in the true edge of a conditional expression, you cannot infer a simple
equivalence as soon as there are or there might be signed zeros involved.