[Bug fortran/19443] New: log10 statement does not work properly in gfortran

[Bug fortran/19443] New: log10 statement does not work properly in gfortran

[mgilbert at airborne dot nrl dot navy dot mil]

The log10 operator in gfortran (cygwin build 20041213) does not work properly if
the statement is operated on a variable, but it works fine when operating on a
constant.  The plain log statement works fine regardless.  If the following
program is executed, one can observer that log for the two methods generates the
same value, but log10 using the different methods produces different results. 
The 10*log10(tst) appears to generate -Infinity, where it should generate -50.

program test
  implicit none
  real :: tst
  
  tst=1e-5
  write(*,*) 10*log(1e-5), 10*log(tst)
  write(*,*) 10*log10(1e-5), 10*log10(tst)
end program test

-- 
           Summary: log10 statement does not work properly in gfortran
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: mgilbert at airborne dot nrl dot navy dot mil
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-14 13:04 -------
This works for me on ppc-darwin, are you sure that is not a bug in cygwin as fortran just calls log10f 
(which might call log10 if your target does not have the C99 math functions).

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[mgilbert at airborne dot nrl dot navy dot mil]

------- Additional Comments From mgilbert at airborne dot nrl dot navy dot mil  2005-01-14 13:20 -------
I tried the exact same code with g95 on cygwin, and it worked flawlessly. 
gfortran must have access to the math library because it is able to evaluate log
properly, right?  How do you suggest that I go about determining whether it is a
cygwin bug or gfortran bug?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-14 13:25 -------
Try this c code:
#include <math.h>
#include <stdio.h>

int main(void)
{
  printf("%f\n", 10*log10(1.0e-5));
  return 0;
}

It might also be the bug that isfinite does not work on cygwin which is a different bug.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[mgilbert at airborne dot nrl dot navy dot mil]

------- Additional Comments From mgilbert at airborne dot nrl dot navy dot mil  2005-01-14 13:34 -------
That c code above worked properly, and running g95 on my test code works fine
also, so I think that pretty much indicates that the issue is with gfortran. 
and gfortran works fine when i do log10(1e-5), but fails on tst=1e-5;
log10(tst).  so it partially works.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-14 13:38 -------
Hmm, try this (if might be that log10f is broken):
#include <math.h>
#include <stdio.h>

int main(void)
{
  float t = 1.0e-5;
  printf("%f\n", 10*log10f(t));
  return 0;
}

If this does not link (try -lm), otherwise I don't know where the bug is.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[mgilbert at airborne dot nrl dot navy dot mil]

------- Additional Comments From mgilbert at airborne dot nrl dot navy dot mil  2005-01-14 13:42 -------
That code works properly also.  I'm downloading an older snapshot of cygwin
gfortran to see if the bug is in there also.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-14 13:51 -------
Then this might be a dup of bug 19363, if both of those C programs worked.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[mgilbert at airborne dot nrl dot navy dot mil]

------- Additional Comments From mgilbert at airborne dot nrl dot navy dot mil  2005-01-14 16:51 -------
My test code works perfectly with the 20041114 cygwin gfortran and earlier
snapshots.

I agree, this could be related to #19363 since the dates of working/non-working
code are similar.  I guess your best bet would be to track down what changes
between 20041114 and 20041212 could cause the errors we're seeing.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[tobi at gcc dot gnu dot org]

------- Additional Comments From tobi at gcc dot gnu dot org  2005-01-17 22:58 -------
(In reply to comment #2)
> I tried the exact same code with g95 on cygwin, and it worked flawlessly. 
> gfortran must have access to the math library because it is able to evaluate log
> properly, right?  How do you suggest that I go about determining whether it is a
> cygwin bug or gfortran bug?

As a data point, I recall that g95 actually implements its own log10 function,
so cygwin might still be broken.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[tobi at gcc dot gnu dot org]

------- Additional Comments From tobi at gcc dot gnu dot org  2005-01-17 23:01 -------
This will teach me not reading all comments. If it did work two months ago it is
indeed more likely that cygwin is not the point of failure.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[paulthomas2 at wanadoo dot fr]

------- Additional Comments From paulthomas2 at wanadoo dot fr  2005-01-25 09:45 -------
I am sorry, I missed this thread completely:

I think that this is a problem with gfortran's log10(real*4).

The following works correctly with the gcc built at the same time as gfortran - 
gcc (GCC) 4.0.0 20050117 (experimental):

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

int main (void)
{
  float tst4 = 1.0e-5;
  double tst8 = 1.0e-5;
  printf ("Literal ....%f\n", 10*log10 (1.0e-5));
  printf ("Float.......%f\n", 10*log10 (tst4));
  printf ("Double......%f\n", 10*log10 (tst8));
  return 0;
}

program test
  implicit none
  REAL*4 :: tst4
  real*8 :: tst8
  tst4 = 1e-5_4
  tst8 = 1e-5_8
  write (*,*) "log with real*4 ......", 10*log (1e-5_4), 10*log (tst4)
  write (*,*) "log with real*8 ......", 10*log (1e-5_8), 10*log (tst8)
  write (*,*) "log10 with real*4 ....", 10*log10 (1e-5_4), 10*log10 (tst4)
  write (*,*) "log10 with real*8 ....", 10*log10 (1e-5_8), 10*log10 (tst8)
end program test

 log with real*4 ......  -115.1292      -115.1292
 log with real*8 ......  -115.129254649702       -115.129254649702
 log10 with real*4 ....  -50.00000          -Infinity
 log10 with real*8 ....  -50.0000000000000       -50.0000000000000

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[sgk at troutmask dot apl dot washington dot edu]

------- Additional Comments From sgk at troutmask dot apl dot washington dot edu  2005-01-31 04:38 -------
Paul,

Your code on i386-*-freebsd produces 

kargl[233] ./log
 log with real*4 ......  -115.1292      -115.1293    
 log with real*8 ......  -115.129254649702       -115.129254649702     
 log10 with real*4 ....  -50.00000      -50.00000    
 log10 with real*8 ....  -50.0000000000000       -50.0000000000000     
kargl[234] gfc -o log -O2 log.f90
kargl[235] ./log
 log with real*4 ......  -115.1292      -115.1293    
 log with real*8 ......  -115.129254649702       -115.129254649702     
 log10 with real*4 ....  -50.00000      -50.00000    
 log10 with real*8 ....  -50.0000000000000       -50.0000000000000     

This is probably target specific.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[tobi at gcc dot gnu dot org]

------- Additional Comments From tobi at gcc dot gnu dot org  2005-02-07 15:10 -------
Can you compile those tests with -fdump-tree-original and look at the
.t02.original dump to determine what functions the C frontend calls, and what
functions the Fortran frontend calls?

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[tobi at gcc dot gnu dot org]

------- Additional Comments From tobi at gcc dot gnu dot org  2005-02-07 15:12 -------
ps it seems that you'll have to use the -fdump-tree-gimple flag for C and look
at the .t08.gimple dump to get something readable.

On i686-pc-linux the C code calls log10, whereas the Fortran code calls
__buitlin_log10.  I would be surprised if the built-in were only broken on some
operating systems, but you'll never know.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[mgilbert at airborne dot nrl dot navy dot mil]

------- Additional Comments From mgilbert at airborne dot nrl dot navy dot mil  2005-03-04 18:01 -------
I updated all cygwin packages recently as well as gfortran before the hd crash
and ensuing lack of cygwin build on gfortran.org.  However, I just got around to
testing the code again.  It appears that everything works correctly now.

  $ gfc --version
  GNU Fortran 95 (GCC 4.0.0 20050131 (experimental))
  Copyright (C) 2005 Free Software Foundation, Inc.

  $ cat log_test.f90
  program test
    implicit none
    REAL*4 :: tst4
    real*8 :: tst8
    tst4 = 1e-5_4
    tst8 = 1e-5_8
    write (*,*) "log with real*4 ......", 10*log (1e-5_4), 10*log (tst4)
    write (*,*) "log with real*8 ......", 10*log (1e-5_8), 10*log (tst8)
    write (*,*) "log10 with real*4 ....", 10*log10 (1e-5_4), 10*log10 (tst4)
    write (*,*) "log10 with real*8 ....", 10*log10 (1e-5_8), 10*log10 (tst8)
  end program test

  $ gfc -o log_test log_test.f90 && ./log_test     
    log with real*4 ......  -115.1292      -115.1292    
    log with real*8 ......  -115.129254649702       -115.129254649702     
    log10 with real*4 ....  -50.00000      -50.00000    
    log10 with real*8 ....  -50.0000000000000       -50.0000000000000  

The "-fdump-tree-gimple" flag doesn't appear to be available for gcc 3.3.3.  For
your reference, here is the gfc .original file

$ gfc -fdump-tree-original log_test.f90
$ cat log_test.f90.t02.orginal
MAIN__ ()
{
  real4 tst4;
  real8 tst8;

  tst4 = 9.99999974737875163555145263671875e-6;
  tst8 = 1.00000000000000008180305391403130954586231382564e-5;
  _gfortran_filename = "log_test.f90";
  _gfortran_line = 7;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  _gfortran_transfer_character ("log with real*4 ......", 22);
  {
    real4 C.452 = -1.1512924957275390625e+2;

    _gfortran_transfer_real (&C.452, 4);
  }
  {
    real4 D.453;

    D.453 = __builtin_logf (tst4) * 1.0e+1;
      _gfortran_transfer_real (&D.453, 4);
  }
  _gfortran_st_write_done ();
  _gfortran_filename = "log_test.f90";
  _gfortran_line = 8;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  _gfortran_transfer_character ("log with real*8 ......", 22);
  {
    real8 C.454 = -1.15129254649702289725610171444714069366455078125e+2;

    _gfortran_transfer_real (&C.454, 8);
  }
  {
    real8 D.455;

    D.455 = __builtin_log (tst8) * 1.0e+1;
    _gfortran_transfer_real (&D.455, 8);
  }
  _gfortran_st_write_done ();
  _gfortran_filename = "log_test.f90";
  _gfortran_line = 9;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  _gfortran_transfer_character ("log10 with real*4 ....", 22);
  {
    real4 C.456 = -5.0e+1;

    _gfortran_transfer_real (&C.456, 4);
  }
  {
    real4 D.457;

    D.457 = __builtin_log10f (tst4) * 1.0e+1;
    _gfortran_transfer_real (&D.457, 4);
  }
  _gfortran_st_write_done ();
  _gfortran_filename = "log_test.f90";
  _gfortran_line = 10;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  _gfortran_transfer_character ("log10 with real*8 ....", 22);
  {
    real8 C.458 = -5.0e+1;

    _gfortran_transfer_real (&C.458, 8);
  }
  {
    real8 D.459;

    D.459 = __builtin_log10 (tst8) * 1.0e+1;
    _gfortran_transfer_real (&D.459, 8);
  }
  _gfortran_st_write_done ();
}
    

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443

[Bug fortran/19443] log10 statement does not work properly in gfortran

[tobi at gcc dot gnu dot org]

------- Additional Comments From tobi at gcc dot gnu dot org  2005-03-13 17:28 -------
Closed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |WORKSFORME


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19443