[Bug fortran/45624] New: Division by zero compiler error

[Bug fortran/45624] New: Division by zero compiler error

[leftynm at umich dot edu]

I am running Fortran 90 on a MacBook Pro, 10.6.4, with a 2.8 GHz Intel Core 2
Duo.
I use terminal to run the code, and my compiler is gfortran.
When I type gfortran -v this is what it tells me:

Using built-in specs.
Target: i386-apple-darwin8.10.1
Configured with: /tmp/gfortran-20090604/ibin/../gcc/configure
--prefix=/usr/local/gfortran --enable-languages=c,fortran
--with-gmp=/tmp/gfortran-20090604/gfortran_libs --enable-bootstrap
Thread model: posix
gcc version 4.5.0 20090604 (experimental) [trunk revision 148180] (GCC) 


Ok, I hope that is enough information off the bat.  Here is my problem I
encountered in my code, shown here in a most simplified form to focus on the
problem:

PROGRAM test

IMPLICIT NONE
REAL*8, PARAMETER       :: alpha = 0.D0
REAL*8, DIMENSION(5,5)  :: matrix
INTEGER                 :: i

matrix = 0.D0

DO i=1,5
    IF (alpha == 0.D0) THEN
        matrix(i,i) = 0.D0
    ELSE
        matrix(i,i) = 1.D0/alpha
    ENDIF
ENDDO

END PROGRAM


When I compile, I get the following:
$ gfortran test.F90 

test.F90:14.26:

        matrix(i,i) = 1.D0/alpha
                          1
Error: Division by zero at (1)


Yes alpha is zero, and yes that line shows me dividing by it, however, I only
divide by this number when it is non-zero.  In my code, most of the time this
number is non-zero. When it is zero, I treat it with an IF-THEN-ELSE type
scenario that does not divide by zero.  The compiler doesn't realize this.

I've tried other variants (such as ELSEIF or switching the order, or using
IF-statements instead of IF-blocks), but they don't seem to matter.

I tried searching for this problem in forums, and I did read the through
gfortran manual (to avoid a RTFM answer).  However, I wasn't able to find a
similar problem or an option to fix this.  

It may seem that not making the variable alpha a PARAMETER will fix the
problem.  However, this isn't an option because of how the problem manifests
within my code.  I need to set this variable as a parameter because of how it
is used elsewhere and passed between modules and subroutines (of course it
doesn't matter in the toy program I showed here).

Thanks for your time.


-- 
           Summary: Division by zero compiler error
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: leftynm at umich dot edu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45624

[Bug fortran/45624] Division by zero compiler error

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2010-09-09 22:19 -------
PARAMETER are special as it is an exact replacement for those variables.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|major                       |normal


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45624

[Bug fortran/45624] Division by zero compiler error

[kargl at gcc dot gnu dot org]

------- Comment #2 from kargl at gcc dot gnu dot org  2010-09-09 22:25 -------
There is no way to fix this problem unless you
would like +inf along the diagonal.

gfortran will constant fold 1./alpha if alpha has
the parameter attribute.  After all, this attribute
tells the compiler that alpha is a named constant,
ie., it will not change.  

PS:  Fortran bugs are never given a severity high
than normal unless the bug breaks bootstrapping
the compiler.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45624

[Bug fortran/45624] Division by zero compiler error

[leftynm at umich dot edu]

------- Comment #3 from leftynm at umich dot edu  2010-09-09 22:56 -------
Ok, if the fact that the variable is a parameter causes this error, I'll have
to adjust the rest of my code accordingly to work around this problem.  It
seems like there should be a better way to fix this, since I want to protect
this variable.
But, I'll take it.  Thanks for the quick reply.


-- 

leftynm at umich dot edu changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45624

[Bug fortran/45624] Division by zero compiler error

[dominiq at lps dot ens dot fr]

------- Comment #4 from dominiq at lps dot ens dot fr  2010-09-09 23:20 -------
You can use the option -fno-range-check. However, the code itself and the
sentence "since I want to protect this variable" in comment #3 let me suspect
that you have not understood what PARAMETER is for: a variable with the
PARAMETER "attribute" is an alias for its value and exists only during the
compilation. 

What you seem to want is the attribute PROTECTED, i.e. a variable that can be
changed only in a specified place as in (F2003 standard draft):

An example of the PROTECTED attribute:

MODULE temperature
REAL, PROTECTED :: temp_c, temp_f
CONTAINS
SUBROUTINE set_temperature_c(c)
REAL, INTENT(IN) :: c
temp_c = c
temp_f = temp_c*(9.0/5.0) + 32
END SUBROUTINE
END MODULE

The PROTECTED attribute ensures that the variables temp_c and temp_f cannot be
modi\fed other than
via the set_temperature_c procedure, thus keeping them consistent with each
other.

The PROTECTED attribute is a f2003 addition, implemented at least from gfortran
4.4.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45624

[Bug fortran/45624] Division by zero compiler error

[leftynm at umich dot edu]

------- Comment #5 from leftynm at umich dot edu  2010-09-10 16:06 -------
Thanks guys.  Yeah, I guess my use of PARAMETER wasn't consistent with how it
works.  I was using it to set a variable such that it cannot be changed.  I
found a work around though lets me keep it as a PARAMETER, but allows me to
compile.

PROGRAM test

IMPLICIT NONE
REAL*8, PARAMETER       :: alpha = 0.D0
REAL*8, DIMENSION(5,5)  :: matrix
INTEGER                 :: i
REAL*8                  :: dummy

matrix = 0.D0

DO i=1,5
    IF (alpha == 0.D0) THEN
        matrix(i,i) = 0.D0
    ELSE
        dummy = alpha
        matrix(i,i) = 1.D0/dummy
    ENDIF
ENDDO

END PROGRAM

This is what I need, since when I don't make my variable a PARAMETER, it causes
other problems.  This requires less work to fix, so I'm happy with it.  I tried
PROTECTED, but my code has to be F90 because it's a small part of a bigger
project.

Thanks to everyone though.  I wouldn't have figured this out without your
inputs.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45624