[Bug c++/41094] New: Erroneous optimization of pow() with -ffast-math

[Bug c++/41094] New: Erroneous optimization of pow() with -ffast-math

[warp at iki dot fi]

Consider the following program:

//----------------------------------------------------------------------
#include <iostream>
#include <cmath>

int main()
{
    for(double p = .25; p <= .5; p += .25)
        for(double x = -10.0; x <= 10.0; x += 20.0)
        {
            std::cout
                << "\nx*x = " << x*x
                << "\npow(100, " << p << ") = " << std::pow(100.0, p)
                << "\npow(x*x, " << p << ") = " << std::pow(x*x, p)
                << "\n";
        }
}
//----------------------------------------------------------------------

When compiling without -ffast-math, the output is the expected one.
However, if compiled with -ffast-math (other optimization options don't
seem to matter), it produces incorrectly "nan" and "-10" for the third
line ("std::pow(x*x, p") when 'x' is negative.

If the -fno-unsafe-math-optimizations option is specified in addition
to -ffast-math, the bug is not triggered.

If "double xx = x*x;" is added inside the loop and then 'xx' is used
instead of 'x*x', the bug is not triggered either.


-- 
           Summary: Erroneous optimization of pow() with -ffast-math
           Product: gcc
           Version: 4.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: warp at iki dot fi
 GCC build triplet: i586-suse-linux
  GCC host triplet: i586-suse-linux
GCC target triplet: i586-suse-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41094

[Bug c++/41094] Erroneous optimization of pow() with -ffast-math

[paolo dot carlini at oracle dot com]

------- Comment #1 from paolo dot carlini at oracle dot com  2009-08-17 18:38 -------
Note, this isn't C++ specific:

int main()
{
  double x = -10.0;
  __builtin_printf("%g\n", __builtin_pow(x * x, 0.25));
}

Richard, can you have a look? It's hot here ;) but first blush I don't see why
fast-math is so special here...


-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41094

[Bug c++/41094] Erroneous optimization of pow() with -ffast-math

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2009-08-18 09:20 -------
Looks like a bug in fold.


-- 

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|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2009-08-18 09:20:14
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41094

[Bug c++/41094] Erroneous optimization of pow() with -ffast-math

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2009-08-18 13:40 -------
Subject: Bug 41094

Author: rguenth
Date: Tue Aug 18 13:40:18 2009
New Revision: 150874

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=150874
Log:
2009-08-18  Richard Guenther  <rguenther@suse.de>

        PR middle-end/41094
        * builtins.c (fold_builtin_pow): Fold pow(pow(x,y),z) to
        pow(x,y*z) only if x is nonnegative.

        * gcc.dg/torture/pr41094.c: New testcase.
        * gcc.dg/torture/builtin-power-1.c: Adjust.
        * gcc.dg/builtins-10.c: Likewise.

Added:
    trunk/gcc/testsuite/gcc.dg/torture/pr41094.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/builtins.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.dg/builtins-10.c
    trunk/gcc/testsuite/gcc.dg/torture/builtin-power-1.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41094

[Bug c++/41094] Erroneous optimization of pow() with -ffast-math

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2009-08-24 09:52 -------
Subject: Bug 41094

Author: rguenth
Date: Mon Aug 24 09:52:03 2009
New Revision: 151051

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=151051
Log:
2009-08-24  Richard Guenther  <rguenther@suse.de>

        PR middle-end/41094
        * builtins.c (fold_builtin_pow): Fold pow(pow(x,y),z) to
        pow(x,y*z) only if x is nonnegative.

        * gcc.dg/torture/pr41094.c: New testcase.
        * gcc.dg/torture/builtin-power-1.c: Adjust.
        * gcc.dg/builtins-10.c: Likewise.

Added:
    branches/gcc-4_4-branch/gcc/testsuite/gcc.dg/torture/pr41094.c
Modified:
    branches/gcc-4_4-branch/gcc/ChangeLog
    branches/gcc-4_4-branch/gcc/builtins.c
    branches/gcc-4_4-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_4-branch/gcc/testsuite/gcc.dg/builtins-10.c
    branches/gcc-4_4-branch/gcc/testsuite/gcc.dg/torture/builtin-power-1.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41094

[Bug c++/41094] Erroneous optimization of pow() with -ffast-math

[rguenth at gcc dot gnu dot org]

------- Comment #5 from rguenth at gcc dot gnu dot org  2009-08-24 09:56 -------
Subject: Bug 41094

Author: rguenth
Date: Mon Aug 24 09:56:30 2009
New Revision: 151052

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=151052
Log:
2009-08-24  Richard Guenther  <rguenther@suse.de>

        PR middle-end/41094
        * builtins.c (fold_builtin_pow): Fold pow(pow(x,y),z) to
        pow(x,y*z) only if x is nonnegative.

        * gcc.dg/torture/pr41094.c: New testcase.
        * gcc.dg/torture/builtin-power-1.c: Adjust.
        * gcc.dg/builtins-10.c: Likewise.

Added:
    branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/torture/pr41094.c
Modified:
    branches/gcc-4_3-branch/gcc/ChangeLog
    branches/gcc-4_3-branch/gcc/builtins.c
    branches/gcc-4_3-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/builtins-10.c
    branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/torture/builtin-power-1.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41094

[Bug c++/41094] Erroneous optimization of pow() with -ffast-math

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2009-08-24 10:03 -------
Fixed.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
      Known to fail|                            |4.3.4 4.4.1
      Known to work|                            |4.3.5 4.4.2 4.5.0
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41094