[gfortran] patch for PR 15292 missing round and roundf

[gfortran] patch for PR 15292 missing round and roundf

[Bud Davis]

adds steve kargl's implmentation of these two functions along with a
configure test.

i expect we will have a few more of these, thus the name
"c99_functions.c".


--bud



Change Log:

2004  Bud Davis  <bdavis9659@comcast.net>
      Steve Kargl  <sgk@troutmask.apl.washington.edu>

	PR gfortran/15292
	* intrinsics/c99_functions.c: New file.
	* Makefile.am: Add new file.
	* configure.ac: Added test for round/roundf.
	* Makefile.in: Regenerate.
	* configure: Regenerate.
        * configure.h.in: Regenerate.

 
Index: gcc/libgfortran/Makefile.am
===================================================================
RCS file: /cvs/gcc/gcc/libgfortran/Makefile.am,v
retrieving revision 1.7
diff -c -3 -n -p -r1.7 Makefile.am
*** gcc/libgfortran/Makefile.am	12 Jun 2004 13:42:45 -0000	1.7
--- gcc/libgfortran/Makefile.am	12 Jun 2004 15:31:11 -0000
*************** gfor_helper_src= \
*** 37,42 ****
--- 37,43 ----
  intrinsics/associated.c \
  intrinsics/abort.c \
  intrinsics/args.c \
+ intrinsics/c99_functions.c \
  intrinsics/cpu_time.c \
  intrinsics/cshift0.c \
  intrinsics/eoshift0.c \
Index: gcc/libgfortran/configure.ac
===================================================================
RCS file: /cvs/gcc/gcc/libgfortran/configure.ac,v
retrieving revision 1.1
diff -c -3 -n -p -r1.1 configure.ac
*** gcc/libgfortran/configure.ac	30 May 2004 21:58:10 -0000	1.1
--- gcc/libgfortran/configure.ac	12 Jun 2004 15:31:14 -0000
*************** AC_CHECK_LIB([mx],[csin],[need_math="no"
*** 166,171 ****
--- 166,175 ----
  # Check for library functions.
  AC_CHECK_FUNCS(getrusage times)
  
+ # Check for some C99 functions
+ AC_CHECK_LIB([m],[round],[AC_DEFINE([HAVE_ROUND],[1],["c99 function"])])
+ AC_CHECK_LIB([m],[roundf],[AC_DEFINE([HAVE_ROUNDF],[1],["c99 function"])])
+ 
  # Let the user override this
  AC_ARG_ENABLE(cmath,
  	AC_HELP_STRING([--enable-cmath],[Include complex math functions]),


and the file:

/* Implementation of various C99 functions 
   Copyright (C) 2004 Free Software Foundation, Inc.

This file is part of the GNU Fortran 95 runtime library (libgfortran).

Libgfortran is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.

Libgfortran is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public
License along with libgfortran; see the file COPYING.LIB.  If not,
write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA.  */

#include "config.h"
#include <sys/types.h>
#include <math.h>
#include "libgfortran.h"


/* algorithm by Steven G. Kargl */

#ifndef HAVE_ROUND
/* round(x)
 * round to nearest integral value.  If the argument is halfway between two
 * integral values then round away from zero.
 */

double
round(double x)
{
   double t;
   int i;

   i = fpclassify(x);
   if (i == FP_INFINITE || i == FP_NAN)
     return (x);

   if (x >= 0.0) 
    {
      t = ceil(x);
      if (t - x > 0.5)
        t -= 1.0;
      return (t);
    } 
   else 
    {
      t = ceil(-x);
      if (t + x > 0.5)
        t -= 1.0;
      return (-t);
    }
}
#endif

#ifndef HAVE_ROUNDF
/* roundf(x)
 * Round to nearest integral value.  If the argument is halfway between two
 * integral values then round away from zero.
 */

float
roundf(float x)
{
   float t;
   int i;

   i = fpclassify(x);
   if (i == FP_INFINITE || i == FP_NAN)
     return (x);

   if (x >= 0.0) 
    {
      t = ceilf(x);
      if (t - x > 0.5)
        t -= 1.0;
      return (t);
    } 
   else 
    {
      t = ceilf(-x);
      if (t + x > 0.5)
        t -= 1.0;
      return (-t);
    }
}
#endif

Re: [gfortran] patch for PR 15292 missing round and roundf

[Paul Brook]

> 2004  Bud Davis  <bdavis9659@comcast.net>
>       Steve Kargl  <sgk@troutmask.apl.washington.edu>
>
> 	PR gfortran/15292
> 	* intrinsics/c99_functions.c: New file.
> 	* Makefile.am: Add new file.
> 	* configure.ac: Added test for round/roundf.
> 	* Makefile.in: Regenerate.
> 	* configure: Regenerate.
>         * configure.h.in: Regenerate.

Applied, thanks.

Paul

Re: [gfortran] patch for PR 15292 missing round and roundf

[Gerald Pfeifer]

On Sat, 12 Jun 2004, Paul Brook wrote:
>> 2004  Bud Davis  <bdavis9659@comcast.net>
>>       Steve Kargl  <sgk@troutmask.apl.washington.edu>
>>
>> 	PR gfortran/15292
>> 	* intrinsics/c99_functions.c: New file.
>> 	* Makefile.am: Add new file.
>> 	* configure.ac: Added test for round/roundf.
>> 	* Makefile.in: Regenerate.
>> 	* configure: Regenerate.
>>         * configure.h.in: Regenerate.
> Applied, thanks.

I'm afraid this is responsible for the following, new bootstrap failure on
i386-unknown-freebsd4.9.  Would you mind having a look?

/files/pfeifer/OBJ-0613-0308/gcc/xgcc -B/files/pfeifer/OBJ-0613-0308/gcc/ -B/sw/gcc-current/i386-unknown-freebsd4.9/bin/ -B/sw/gcc-current/i386-unknown-freebsd4.9/lib/ -isystem /sw/gcc-current/i386-unknown-freebsd4.9/include -isystem /sw/gcc-current/i386-unknown-freebsd4.9/sys-include -DHAVE_CONFIG_H -I. -I/sw/test/gcc/cvs/libgfortran -I. -I/sw/test/gcc/cvs/libgfortran/io -O2 -g -O2 -std=gnu99 -O2 -g -O2 -c /sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c  -fPIC -DPIC -o .libs/c99_functions.o
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c: In function `round':
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c:39: warning: implicit declaration of function 'fpclassify'
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c:40: error: 'FP_INFINITE' undeclared (first use in this function)
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c:40: error: (Each undeclared identifier is reported only once
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c:40: error: for each function it appears in.)
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c:40: error: 'FP_NAN' undeclared (first use in this function)
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c: In function `roundf':
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c:71: error: 'FP_INFINITE' undeclared (first use in this function)
/sw/test/gcc/cvs/libgfortran/intrinsics/c99_functions.c:71: error: 'FP_NAN' undeclared (first use in this function)
gmake[3]: *** [c99_functions.lo] Error 1
gmake[3]: Leaving directory `/files/pfeifer/OBJ-0613-0308/i386-unknown-freebsd4.9/libgfortran'
gmake[2]: *** [all] Error 2
gmake[2]: Leaving directory `/files/pfeifer/OBJ-0613-0308/i386-unknown-freebsd4.9/libgfortran'
gmake[1]: *** [all-target-libgfortran] Error 2

Gerald
-- 
Gerald Pfeifer (Jerry)   gerald@pfeifer.com   http://www.pfeifer.com/gerald/

Re: [gfortran] patch for PR 15292 missing round and roundf

[Steve Kargl]

On Sun, Jun 13, 2004 at 06:04:07PM +0200, Gerald Pfeifer wrote:
> On Sat, 12 Jun 2004, Paul Brook wrote:
> >> 2004  Bud Davis  <bdavis9659@comcast.net>
> >>       Steve Kargl  <sgk@troutmask.apl.washington.edu>
> >>
> >> 	PR gfortran/15292
> >> 	* intrinsics/c99_functions.c: New file.
> >> 	* Makefile.am: Add new file.
> >> 	* configure.ac: Added test for round/roundf.
> >> 	* Makefile.in: Regenerate.
> >> 	* configure: Regenerate.
> >>         * configure.h.in: Regenerate.
> > Applied, thanks.
> 
> I'm afraid this is responsible for the following, new bootstrap failure on
> i386-unknown-freebsd4.9.  Would you mind having a look?
> 

This is fixed in FreeBSD-5.  Freebsd-current also has round[f]()
because I submitted the implementation which is the same as 
the code that Bud committed.

However, I suspect that other platforms don't supply an fpclassify().
There are a few options.

(1) We can remove the offending code fragment.

    i = fpclassify(x);
    if (i == FP_INFINITE || i == FP_NAN)
       return (x);

    and potentially do arithematic and comparisons involving Inf and NaN.

(2) We can include a implementation of fpclassify().  I don't have 
    such code.

(3) Require platforms to have a C99 environment to compile gfortran.

(4) We can use sysdeps/ieee754/dbl-64/s_round.c and 
    sysdeps/ieee754/flt-32/s_roundf.c from the glic-2.3.2.  However,
    someone will need to clarify a possible copyright infringement.
    Hint compare these files with s_ceil.c and s_ceilf.c from fdlibm.

    


-- 
Steve

Re: [gfortran] patch for PR 15292 missing round and roundf

[Paolo Bonzini]

[-- Attachment #1: Type: text/plain, Size: 311 bytes --]

> (2) We can include a implementation of fpclassify().  I don't have 
>     such code.

if (x != x) can be used to check for NaN.

if (x - x != 0.0) can be used to check for NaN or infinity.

if (x + x == x) can be used to check for infinity or zero.

See the attached test programlet.

Hope this helps,

Paolo

[-- Attachment #2: prova.c --]
[-- Type: text/plain, Size: 779 bytes --]

char * is_nan_test (float x, char *what)
{
  if (x != x)
    return what;
  else
    return "    ";
}

char * is_nan_inf_test (float x, char *what)
{
  if (x - x != 0.0)
    return what;
  else
    return "    ";
}

char * is_inf_zero_test (float x, char *what)
{
  if (x + x == x)
    return what;
  else
    return "    ";
}

void test (float x, char *what)
{
  printf ("%s | %s |  %s   |  %s\n", what,
	  is_nan_test (x, what),
	  is_nan_inf_test (x, what),
	  is_inf_zero_test (x, what));
}

int main()
{
  printf ("     |  NaN |  oo/NaN |  oo/0\n");
  printf ("-----------------------------\n");

  test (0.0, " 0.0");
  test (__builtin_inf(), " +oo");
  test (-__builtin_inf(), " -oo");
  test (__builtin_nan("0x0"), " NaN");
  test (5.0, " 5.0");
  test (-5.0, "-5.0");
}

Re: [gfortran] patch for PR 15292 missing round and roundf

[Jakub Jelinek]

On Mon, Jun 14, 2004 at 12:52:41PM +0200, Paolo Bonzini wrote:
> >(2) We can include a implementation of fpclassify().  I don't have 
> >    such code.
> 
> if (x != x) can be used to check for NaN.
> 
> if (x - x != 0.0) can be used to check for NaN or infinity.
> 
> if (x + x == x) can be used to check for infinity or zero.

Isn't fpclassify supposed not to generate any exceptions though?
The second test will generate invalid exception while fpclassify
will not (plus if one of the arguments is sNaN all 3 will throw
exception while sNaN will not.

	Jakub

Re: [gfortran] patch for PR 15292 missing round and roundf

[Paolo Bonzini]

>>>(2) We can include a implementation of fpclassify().  I don't have 
>>>   such code.
>>
>>if (x != x) can be used to check for NaN.
>>
>>if (x - x != 0.0) can be used to check for NaN or infinity.
>>
>>if (x + x == x) can be used to check for infinity or zero.
> 
> 
> Isn't fpclassify supposed not to generate any exceptions though?
> The second test will generate invalid exception while fpclassify
> will not (plus if one of the arguments is sNaN all 3 will throw
> exception while sNaN will not.

Yes, but Bud was asking in the context of round.  IIRC, round should 
generate an invalid exception if the parameter is a NaN.

Paolo

Re: [gfortran] patch for PR 15292 missing round and roundf

[Gerald Pfeifer]

On Sun, 13 Jun 2004, Steve Kargl wrote:
> (3) Require platforms to have a C99 environment to compile gfortran.

That sounds a bit strict?

And, by the way, it's not straightforward to not compile this: if you
have libgfortran, but not the fortran frontend (gcc/fortran)

  % ls -lad gcc/fortran
  /bin/ls: gcc/fortran: No such file or directory
  % ls -lad libgfortran/
  drwxr-xr-x  8 user group 640 2004-06-14 13:03 libgfortran/

your bootstrap will still fail for --enable-languages=c,c++,objc.

  /tmp/OBJ-0614-1414/gcc/gfortran -B/tmp/OBJ-0614-1414/gcc/
  -B/group/user/gcc-i386/i686-pc-linux-gnu/bin/
  -B/group/user/gcc-i386/i686-pc-linux-gnu/lib/ -isystem
  /group/user/gcc-i386/i686-pc-linux-gnu/include -isystem
  /group/user/gcc-i386/i686-pc-linux-gnu/sys-include -Wall -fno-repack-arrays
  -fno-underscoring -c /cvs/gcc/libgfortran/intrinsics/selected_kind.f90
  -fPIC -DPIC -o .libs/selected_kind.o
  ./libtool: line 1: /tmp/OBJ-0614-1414/gcc/gfortran: No such file or directory
  gmake[3]: *** [selected_kind.lo] Error 1
  gmake[3]: Leaving directory `/tmp/OBJ-0614-1414/i686-pc-linux-gnu/libgfortran'
  gmake[2]: *** [all] Error 2
  gmake[2]: Leaving directory
  `/tmp/OBJ-0614-1414/i686-pc-linux-gnu/libgfortran'
  gmake[1]: *** [all-target-libgfortran] Error 2

Gerald
-- 
Gerald Pfeifer (Jerry)   gerald@pfeifer.com   http://www.pfeifer.com/gerald/

Re: [gfortran] patch for PR 15292 missing round and roundf

[Tobias Schlüter]

Gerald Pfeifer wrote:
> And, by the way, it's not straightforward to not compile this: if you
> have libgfortran, but not the fortran frontend (gcc/fortran)
> 
>   % ls -lad gcc/fortran
>   /bin/ls: gcc/fortran: No such file or directory
>   % ls -lad libgfortran/
>   drwxr-xr-x  8 user group 640 2004-06-14 13:03 libgfortran/
> 
> your bootstrap will still fail for --enable-languages=c,c++,objc.
> 
I believe this is the same failure as PR 15548? This needs someone with 
sufficient rights to fix the CVS.

- Tobi

Re: [gfortran] patch for PR 15292 missing round and roundf

[Gerald Pfeifer]

On Mon, 14 Jun 2004, Tobias Schlüter wrote:
>> your bootstrap will still fail for --enable-languages=c,c++,objc.
> I believe this is the same failure as PR 15548? This needs someone with
> sufficient rights to fix the CVS.

Yes, and no.  Changing the CVSROOT/modules will avoid this problem for
many users (and I'll try to do that tomorrow) but the basic bug still
remains:

If I configure not to build gfortran, libgfortran should not be built.
Never.  Ever.

Gerald
-- 
Gerald Pfeifer (Jerry)   gerald@pfeifer.com   http://www.pfeifer.com/gerald/

Re: [gfortran] patch for PR 15292 missing round and roundf

[Paul Brook]

On Monday 14 June 2004 14:30, Gerald Pfeifer wrote:
> On Mon, 14 Jun 2004, Tobias Schlüter wrote:
> >> your bootstrap will still fail for --enable-languages=c,c++,objc.
> >
> > I believe this is the same failure as PR 15548? This needs someone with
> > sufficient rights to fix the CVS.
>
> Yes, and no.  Changing the CVSROOT/modules will avoid this problem for
> many users (and I'll try to do that tomorrow) but the basic bug still
> remains:
>
> If I configure not to build gfortran, libgfortran should not be built.
> Never.  Ever.

This problem isn't specific to fortran.
For example removing gcc/cp will cause libstdc++-v3 to be built 
unconditionally.

Paul

Re: [gfortran] patch for PR 15292 missing round and roundf

[Paolo Bonzini]

> This problem isn't specific to fortran.
> For example removing gcc/cp will cause libstdc++-v3 to be built 
> unconditionally.

Adding a --enable-libgfortran to the front-end will help.

Paolo

Re: [gfortran] patch for PR 15292 missing round and roundf

[Paolo Bonzini]

> This problem isn't specific to fortran.
> For example removing gcc/cp will cause libstdc++-v3 to be built 
> unconditionally.

Adding a --enable-libgfortran to the front-end will help.

Paolo

Re: [gfortran] patch for PR 15292 missing round and roundf

[David Edelsohn]

	This patch causes other failures on AIX 5.1 because some of the
macros used by the implementation of the new functions are not defined.
AIX 5.1 has most of C99 support.

David


/gcc/dje/build/powerpc-ibm-aix5.1.0.0-20040613/gcc/xgcc -B/gcc/dje/build/powerpc-ibm-aix5.1.0.0-20040613/gcc/ -B/gcc/dje/install/powerpc-ibm-aix5.1.0.0-20040613/powerpc-ibm-aix5.1.0.0/bin/ -B/gcc/dje/install/powerpc-ibm-aix5.1.0.0-20040613/powerpc-ibm-aix5.1.0.0/lib/ -isystem /gcc/dje/install/powerpc-ibm-aix5.1.0.0-20040613/powerpc-ibm-aix5.1.0.0/include -isystem /gcc/dje/install/powerpc-ibm-aix5.1.0.0-20040613/powerpc-ibm-aix5.1.0.0/sys-include -DHAVE_CONFIG_H -I. -I/gcc/dje/src/libgfortran -I. -I/gcc/dje/src/libgfortran/io -O2 -g -O2 -std=gnu99 -O2 -g -O2 -c /gcc/dje/src/libgfortran/intrinsics/c99_functions.c   -DPIC -o .libs/c99_functions.o
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c: In function `round':
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:39: warning: implicit declaration of function 'fpclassify'
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:40: error: 'FP_INFINITE' undeclared (first use in this function)
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:40: error: (Each undeclared identifier is reported only once
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:40: error: for each function it appears in.)
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:40: error: 'FP_NAN' undeclared (first use in this function)
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c: In function `roundf':
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:71: error: 'FP_INFINITE' undeclared (first use in this function)
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:71: error: 'FP_NAN' undeclared (first use in this function)
/gcc/dje/src/libgfortran/intrinsics/c99_functions.c:76: warning: implicit declaration of function 'ceilf'