[Bug libfortran/26253] New: FAIL: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90 execution

[Bug libfortran/26253] New: FAIL: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90 execution

[danglin at gcc dot gnu dot org]

Executing on host: /users/bin/gcc/gcc-4.1/objdir/gcc/testsuite/../gfortran
-B/us
ers/bin/gcc/gcc-4.1/objdir/gcc/testsuite/../
/users/bin/gcc/gcc-4.1/gcc/gcc/test
suite/gfortran.fortran-torture/execute/intrinsic_set_exponent.f90  -w  -O0  
-L/
users/bin/gcc/gcc-4.1/objdir/hppa1.1-hp-hpux10.20/./libgfortran/.libs
-L/users/b
in/gcc/gcc-4.1/objdir/hppa1.1-hp-hpux10.20/./libgfortran/.libs
-L/users/bin/gcc/
gcc-4.1/objdir/hppa1.1-hp-hpux10.20/./libiberty  -lm   -o
/users/bin/gcc/gcc-4.1
/objdir/gcc/testsuite/intrinsic_set_exponent.x    (timeout = 600)
PASS: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90 compilation,
-O0
Setting LD_LIBRARY_PATH to
.:/users/bin/gcc/gcc-4.1/objdir/hppa1.1-hp-hpux10.20/
./libgfortran/.libs:/users/bin/gcc/gcc-4.1/objdir/hppa1.1-hp-hpux10.20/./libgfor
tran/.libs:/users/bin/gcc/gcc-4.1/objdir/gcc:.:/users/bin/gcc/gcc-4.1/objdir/hpp
a1.1-hp-hpux10.20/./libgfortran/.libs:/users/bin/gcc/gcc-4.1/objdir/hppa1.1-hp-h
pux10.20/./libgfortran/.libs:/users/bin/gcc/gcc-4.1/objdir/gcc
FAIL: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90 execution, 
-O
0

Breakpoint 1, _gfortran_abort ()
    at ../../../gcc/libgfortran/intrinsics/abort.c:38
38        close_units ();
(gdb) bt
#0  _gfortran_abort () at ../../../gcc/libgfortran/intrinsics/abort.c:38
#1  0x3e5c in test_real8_ ()
    at
/users/bin/gcc/gcc-4.1/gcc/gcc/testsuite/gfortran.fortran-torture/execute/intrinsic_set_exponent.f90:59
#2  0x39e4 in MAIN__ ()
    at
/users/bin/gcc/gcc-4.1/gcc/gcc/testsuite/gfortran.fortran-torture/execute/intrinsic_set_exponent.f90:5
#3  0x42bc in main (argc=1, argv=0x7b03b054)
    at ../../../gcc/libgfortran/fmain.c:18
#4  0x7af1d86c in ?? () from /usr/lib/libc.1
(gdb) break intrinsic_set_exponent.f90:59
...
Breakpoint 2, test_real8_ ()
    at
/users/bin/gcc/gcc-4.1/gcc/gcc/testsuite/gfortran.fortran-torture/execute/intrinsic_set_exponent.f90:59
59        if (y .ne. 128.0) call abort()
(gdb) p y
$1 = 128.00000000000003

It looks like the rounding error arises from the call to pow in scalbn.
This is the return value:

(gdb) info reg fr4
fr4     (single precision)     3.75
fr4     (double precision)     256.00000000000006

Not sure why, but the test doesn't fail when linked with -static.  I also
can't duplicate with a simple test program.


-- 
           Summary: FAIL: gfortran.fortran-
                    torture/execute/intrinsic_set_exponent.f90 execution
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libfortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: danglin at gcc dot gnu dot org
 GCC build triplet: hppa1.1-hp-hpux10.20
  GCC host triplet: hppa1.1-hp-hpux10.20
GCC target triplet: hppa1.1-hp-hpux10.20


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] FAIL: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90 execution

[fxcoudert at gcc dot gnu dot org]

------- Comment #1 from fxcoudert at gcc dot gnu dot org  2006-02-14 11:19 -------
This reminds me of PR25370... (for the -static effect).

I'm currently bootstrapping mainline on a hppa2.0w-hp-hpux11.23 (thanks to HP
testdrive) to nail down this one and PR26252. John, are those the only
hppa-hpux failures you see in the testsuite?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] FAIL: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90 execution

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #2 from dave at hiauly1 dot hia dot nrc dot ca  2006-02-14 14:15 -------
Subject: Re:  FAIL: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90
execution

> This reminds me of PR25370... (for the -static effect).
> 
> I'm currently bootstrapping mainline on a hppa2.0w-hp-hpux11.23 (thanks to HP
> testdrive) to nail down this one and PR26252. John, are those the only
> hppa-hpux failures you see in the testsuite?

Here are the failures that I see on hppa1.1-hp-hpu10.20:
http://gcc.gnu.org/ml/gcc-testresults/2006-02/msg00635.html

I don't see any failures at the moment on 4.1.0 on hppa2.0w-hp-hpux11.11.
See http://gcc.gnu.org/ml/gcc-testresults/2006-02/msg00637.html

hpux10.20 doesn't have scalbn and hpux11.x does, so that's one difference.
The problem appears to be with the pow function called from the libgfortran
version of scalbn.  I had looked at the arguments in the call to pow
in scalbn and they appear fine.  However, the return contains a small
rounding error.  I didn't see the same rounding error when I wrote a
little test program to try duplicate the effect.

Possibly, the default rounding mode is somehow getting changed in
the test program.  scalbn is called several times before the error
occurs.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] FAIL: gfortran.fortran-torture/execute/intrinsic_set_exponent.f90 execution

[fxcoudert at gcc dot gnu dot org]

------- Comment #3 from fxcoudert at gcc dot gnu dot org  2006-02-15 09:15 -------
(In reply to comment #2)
> hpux10.20 doesn't have scalbn and hpux11.x does, so that's one difference.

On hppa2.0w-hp-hpux11.23, I see this same bug if I undefine HAVE_SCALBN in
libgfortran/config.h. I'll look into it sometime tonight or tomorrow.


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fxcoudert at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2006-02-15 09:15:24
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[fxcoudert at gcc dot gnu dot org]

------- Comment #4 from fxcoudert at gcc dot gnu dot org  2006-02-17 20:33 -------
This is a problem in libgfortran scalbn fallback version not handling correctly
denormals. A i686-linux testcase is included below. I switch this to severity
"minor".

$ cat a.c
#include <stdio.h>
#include <math.h>

double
gfortran_scalbn(double x, int y)
{
  static double two = 2;
  return x * pow(two, y);
}

int main (void)
{
  double x, y;
  int n, dummy;

  n = 1024;
  x = 2.22507385850720088902458687608585988765042311224e-308;
  y = gfortran_scalbn (frexp (x, &dummy), n);
  printf ("%lg\n", y);
  return 0;
}

$ gcc a.c -lm && ./a.out
inf

If we use the libm scalbn instead (at least on i686-linux), the result printed
is 1.79769e+308.


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |minor
  GCC build triplet|hppa1.1-hp-hpux10.20        |
   GCC host triplet|hppa1.1-hp-hpux10.20        |
 GCC target triplet|hppa1.1-hp-hpux10.20        |
   Last reconfirmed|2006-02-15 09:15:24         |2006-02-17 20:33:15
               date|                            |
            Summary|FAIL: gfortran.fortran-     |fallback scalbn doesn't
                   |torture/execute/intrinsic_se|handle denormals correctly
                   |t_exponent.f90 execution    |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[danglin at gcc dot gnu dot org]

------- Comment #5 from danglin at gcc dot gnu dot org  2007-01-07 00:55 -------
This failure also occurs in 4.0.3 hppa1.1-hp-hpux10.20.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[fxcoudert at gcc dot gnu dot org]

------- Comment #6 from fxcoudert at gcc dot gnu dot org  2007-09-19 09:16 -------
For some reason, I had completely forgotten about ldexp(). But as it is C89, it
should be available, and as all GCC supported targets now have (FLT_RADIX ==
2), it should fix the problem. John, could you confirm that the following patch
fixes the testsuite failure?



Index: libgfortran/intrinsics/c99_functions.c
===================================================================
--- libgfortran/intrinsics/c99_functions.c      (revision 128540)
+++ libgfortran/intrinsics/c99_functions.c      (working copy)
@@ -341,7 +341,11 @@ log10f(float x)
 double
 scalbn(double x, int y)
 {
+#if (FLT_RADIX == 2) && defined(HAVE_LDEXP)
+  return ldexp (x, y);
+#else
   return x * pow(FLT_RADIX, y);
+#endif
 }
 #endif


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |fxcoudert at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
           Keywords|                            |patch
   Last reconfirmed|2006-02-17 20:33:15         |2007-09-19 09:16:22
               date|                            |
   Target Milestone|---                         |4.3.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #7 from dave at hiauly1 dot hia dot nrc dot ca  2007-09-20 02:24 -------
Subject: Re:  fallback scalbn doesn't handle denormals correctly

> For some reason, I had completely forgotten about ldexp(). But as it is C89, it
> should be available, and as all GCC supported targets now have (FLT_RADIX ==
> 2), it should fix the problem. John, could you confirm that the following patch
> fixes the testsuite failure?

Testing is underway but this looks close to obvious.  From the HP-UX 11.11
manpage for scalbn:

      The scalbn() function returns x * r**n, where r is the radix of the
      machine's floating-point arithmetic.  When r is 2 (as it is on all
      PA-RISC systems), scalbn() is equivalent to ldexp().

Thanks,
Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[fxcoudert at gcc dot gnu dot org]

------- Comment #8 from fxcoudert at gcc dot gnu dot org  2007-09-21 10:54 -------
Subject: Bug 26253

Author: fxcoudert
Date: Fri Sep 21 10:54:20 2007
New Revision: 128648

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=128648
Log:
        PR libfortran/26253
        * intrinsics/c99_functions.c (scalbn): Use ldexp if appopriate.

Modified:
    trunk/libgfortran/ChangeLog
    trunk/libgfortran/intrinsics/c99_functions.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[fxcoudert at gcc dot gnu dot org]

------- Comment #9 from fxcoudert at gcc dot gnu dot org  2007-09-21 10:55 -------
This should now be fixed. Please reopen if I fumbled in any way and the issue
is still present.


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[danglin at gcc dot gnu dot org]

------- Comment #10 from danglin at gcc dot gnu dot org  2008-01-21 17:25 -------
Subject: Bug 26253

Author: danglin
Date: Mon Jan 21 17:24:30 2008
New Revision: 131698

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=131698
Log:
        PR libfortran/34699
        Backport:
        2007-09-21  Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>

        PR libfortran/26253
        * intrinsics/c99_functions.c (scalbn): Use ldexp if appopriate.


Modified:
    branches/gcc-4_2-branch/libgfortran/ChangeLog
    branches/gcc-4_2-branch/libgfortran/intrinsics/c99_functions.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253

[Bug libfortran/26253] fallback scalbn doesn't handle denormals correctly

[danglin at gcc dot gnu dot org]

------- Comment #11 from danglin at gcc dot gnu dot org  2008-01-21 18:13 -------
Subject: Bug 26253

Author: danglin
Date: Mon Jan 21 18:13:04 2008
New Revision: 131699

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=131699
Log:
        PR libfortran/34699
        Backport:
        2007-09-21  Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>

        PR libfortran/26253
        * intrinsics/c99_functions.c (scalbn): Use ldexp if appopriate.


Modified:
    branches/gcc-4_1-branch/libgfortran/ChangeLog
    branches/gcc-4_1-branch/libgfortran/intrinsics/c99_functions.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26253