[Bug fortran/15205] New: NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[Bug fortran/15205] New: NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[gcc-bugzilla at gcc dot gnu dot org]

The NEAREST intrinsic for DOUBLE PRECISION kind type is returning the
wrong value.  Consider the following program:

  program test_nearest

  implicit none

  real(4) sr, sx, ss
  real(8) dr, dx, ds

  sx = 1.237e0_sp
  ss = 1.e0_sp
  sr = nearest(sx,ss)
  write(*,1) 'nearest(', sx, ',', ss, ') = ', sr
1 format(a,e15.8,a,e15.8,a,e15.8)
  ss = -1.e0_sp
  sr = nearest(sx,ss)
  write(*,1) 'nearest(', sx, ',', ss, ') = ', sr

  dx = 1.237e0_dp
  ds = 1.e0_dp
  dr = nearest(dx,ds)
  write(*,2) 'nearest(', dx, ',', ds, ') = ', dr
2 format(a,e23.16,a,e23.16,a,e23.16)
  ds = -1.e0_dp
  dr = nearest(dx,ds)
  write(*,2) 'nearest(', dx, ',', ds, ') = ', dr

end program test_nearest

Gfortran yields

kargl[328] ./nearest
nearest( 0.12369999E+01, 1.00000000E+00) =  0.12370001E+01
nearest( 0.12369999E+01,-1.00000000E+00) =  0.12369998E+01
nearest( 0.1237000000000000E+01, 1.0000000000000000E+00) = -0.4303249471964870E-36
nearest( 0.1237000000000000E+01,-1.0000000000000000E+00) = -0.4303249471964870E-36

NAG's compiler gives

nearest( 0.12370000E+01, 0.10000000E+01) =  0.12370001E+01
nearest( 0.12370000E+01,-0.10000000E+01) =  0.12369999E+01
nearest( 0.1237000000000000E+01, 0.1000000000000000E+01) =  0.1237000000000000E+01
nearest( 0.1237000000000000E+01,-0.1000000000000000E+01) =  0.1237000000000000E+01

Environment:
System: FreeBSD c-67-168-59-70.client.comcast.net 5.2-CURRENT FreeBSD 5.2-CURRENT #2: Sat Apr 24 10:33:08 PDT 2004 kargl@c-67-168-59-70.client.comcast.net:/usr/obj/usr/src/sys/HOTRATS i386


	
host: i386-unknown-freebsd5.2
build: i386-unknown-freebsd5.2
target: i386-unknown-freebsd5.2
configured with: ../gcc/configure --prefix=/home/kargl/gcc/work --disable-libmudflap --enable-languages=c,f95 --with-gmp=/usr/local

How-To-Repeat:
	See above.

-- 
           Summary: NEAREST intrinsic returns wrong value in DOUBLE
                    PRECISION
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: kargl at c-67-168-59-70 dot client dot comcast dot net
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i386-unknown-freebsd5.2
  GCC host triplet: i386-unknown-freebsd5.2
GCC target triplet: i386-unknown-freebsd5.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|NEAREST intrinsic returns   |[gfortran] NEAREST intrinsic
                   |wrong value in DOUBLE       |returns wrong value in
                   |PRECISION                   |DOUBLE PRECISION
   Target Milestone|---                         |tree-ssa
            Version|unknown                     |tree-ssa


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug libfortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-04-29 18:25 -------
I can confirm this on i686-pc-linux-gnu but works on power-apple-darwin for some reason.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
          Component|fortran                     |libfortran
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-04-29 18:25:25
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug libfortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[bdavis9659 at comcast dot net]

------- Additional Comments From bdavis9659 at comcast dot net  2004-04-30 00:36 -------
for some reason i had to remove the _sp and _dp to get it to compile ?

this is not a library problem, it is in the front end.

[bdavis@localhost pr15205]$ gfortran -static a.f90
[bdavis@localhost pr15205]$ nm a.out |grep near
527:08049140 T _gfortran_nearest_r4

or look at the output of -fdump-tree-all


[bdavis@localhost pr15205]$ grep near a.f90.t03.original
12:  sr = _gfortran_nearest_r4 (sx, ss);
19:  _gfortran_transfer_character ("nearest(", 8);
27:  sr = _gfortran_nearest_r4 (sx, ss);
34:  _gfortran_transfer_character ("nearest(", 8);
43:  dr = (real8)_gfortran_nearest_r4 (dx, ds);
50:  _gfortran_transfer_character ("nearest(", 8);
58:  dr = (real8)_gfortran_nearest_r4 (dx, ds);
65:  _gfortran_transfer_character ("nearest(", 8);


 



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-04-30 16:35 -------
(In reply to comment #2)
> for some reason i had to remove the _sp and _dp to get it to compile ?
> 
> this is not a library problem, it is in the front end.
> 
> [bdavis@localhost pr15205]$ gfortran -static a.f90
> [bdavis@localhost pr15205]$ nm a.out |grep near
> 527:08049140 T _gfortran_nearest_r4
> 
> or look at the output of -fdump-tree-all

-fdump-parse-tree can be very helpful as well (after removing _sp and replacing
_dp by _8):

[tobi@marktplatz tests]$ gfortran pr15205.f90 -fdump-parse-tree
 
        Namespace: A-Z: (UNKNOWN 0)
        symtree: dr  Ambig 0
        symbol dr (REAL 8)(VARIABLE UNKNOWN-INTENT UNKNOWN-ACCESS UNKNOWN-PROC)
 
        symtree: sx  Ambig 0
        symbol sx (REAL 4)(VARIABLE UNKNOWN-INTENT UNKNOWN-ACCESS UNKNOWN-PROC)
 
        symtree: ds  Ambig 0
        symbol ds (REAL 8)(VARIABLE UNKNOWN-INTENT UNKNOWN-ACCESS UNKNOWN-PROC)
 
        symtree: sr  Ambig 0
        symbol sr (REAL 4)(VARIABLE UNKNOWN-INTENT UNKNOWN-ACCESS UNKNOWN-PROC)
 
        symtree: dx  Ambig 0
        symbol dx (REAL 8)(VARIABLE UNKNOWN-INTENT UNKNOWN-ACCESS UNKNOWN-PROC)
 
        symtree: nearest  Ambig 0
        symbol nearest (UNKNOWN 0)(PROCEDURE UNKNOWN-INTENT UNKNOWN-ACCESS
UNKNOWN-PROC FUNCTION)
        result: nearest
 
        symtree: ss  Ambig 0
        symbol ss (REAL 4)(VARIABLE UNKNOWN-INTENT UNKNOWN-ACCESS UNKNOWN-PROC)
 
        symtree: test_nearest  Ambig 0
        symbol test_nearest (UNKNOWN 0)(PROGRAM UNKNOWN-INTENT UNKNOWN-ACCESS
UNKNOWN-PROC)
 
 
      ASSIGN sx 0.1237e1
      ASSIGN ss 0.1e1
      ASSIGN sr _gfortran_nearest[[((sx) (ss))]]
      WRITE UNIT=6 FMT=1
      TRANSFER 'nearest('
      TRANSFER sx
      TRANSFER ','
      TRANSFER ss
      TRANSFER ') = '
      TRANSFER sr
      DT_END
      ASSIGN ss -0.1e1
      ASSIGN sr _gfortran_nearest[[((sx) (ss))]]
      WRITE UNIT=6 FMT=1
      TRANSFER 'nearest('
      TRANSFER sx
      TRANSFER ','
      TRANSFER ss
      TRANSFER ') = '
      TRANSFER sr
      DT_END
      ASSIGN dx 0.1237e1_8
      ASSIGN ds 0.1e1_8
      ASSIGN dr __convert_r4_r8[[((_gfortran_nearest[[((dx) (ds))]]))]]
      WRITE UNIT=6 FMT=2
      TRANSFER 'nearest('
      TRANSFER dx
      TRANSFER ','
      TRANSFER ds
      TRANSFER ') = '
      TRANSFER dr
      DT_END
      ASSIGN ds -0.1e1_8
      ASSIGN dr __convert_r4_r8[[((_gfortran_nearest[[((dx) (ds))]]))]]
      WRITE UNIT=6 FMT=2
      TRANSFER 'nearest('
      TRANSFER dx
      TRANSFER ','
      TRANSFER ds
      TRANSFER ') = '
      TRANSFER dr
      DT_END
 
[tobi@marktplatz tests]$

So yeah, the frontend is at fault.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-04-30 16:44 -------
I have a fix for this, as usual it's waiting for my copyright assignment. It
consists in adding a resolution function for the NEAREST intrinsic, so that the
type of the argument is properly propagated.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |Tobias dot Schlueter at
                   |                            |physik dot uni-muenchen dot
                   |                            |de


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-04-30 17:07 -------
I forgot to include this: with my patch Steve's program gives the correct answer
(except for the rounding, I don't know if that's another bug):
[tobi@marktplatz tests]$ LD_LIBRARY_PATH=~tobi/usr/lib ./a.out
nearest( 0.12369999E+01, 1.00000000E+00) =  0.12370001E+01
nearest( 0.12369999E+01,-1.00000000E+00) =  0.12369998E+01
nearest( 0.1237000000000000E+01, 1.0000000000000000E+00) =  0.1237000000000000E+01
nearest( 0.1237000000000000E+01,-1.0000000000000000E+00) =  0.1236999999999999E+01


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[bdavis at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |Tobias dot Schlueter at
                   |dot org                     |physik dot uni-muenchen dot
                   |                            |de
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-05-08 16:17 -------
Patch here: http://gcc.gnu.org/ml/fortran/2004-05/msg00063.html

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-08 17:28 -------
Now I understand why it works on PPC because on PPC single pression fp are passed by register which 
are always in double pression.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch, wrong-code


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-05-14 13:51 -------
Subject: Bug 15205

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	tobi@gcc.gnu.org	2004-05-14 13:51:27

Modified files:
	gcc/fortran    : ChangeLog intrinsic.c intrinsic.h iresolve.c 

Log message:
	PR fortran/15205
	* iresolve.c (gfc_resolve_nearest): Add new function.
	* intrinsic.h: ... declare it here.
	* intrinsic.c (add_functions): ... add it as resolving function
	for NEAREST.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/ChangeLog.diff?cvsroot=gcc&r1=1.9&r2=1.10
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/intrinsic.c.diff?cvsroot=gcc&r1=1.3&r2=1.4
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/intrinsic.h.diff?cvsroot=gcc&r1=1.3&r2=1.4
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/iresolve.c.diff?cvsroot=gcc&r1=1.3&r2=1.4



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205

[Bug fortran/15205] [gfortran] NEAREST intrinsic returns wrong value in DOUBLE PRECISION

[tobi at gcc dot gnu dot org]

------- Additional Comments From tobi at gcc dot gnu dot org  2004-05-14 14:00 -------
Fixed by the above commit.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15205