etime

[François-Xavier Coudert]

Hello,

I am trying to use gfortran to compile scientific code (including BLAS,
LAPACK and ARPACK packages), but my compilation (using gfortran-20040922)
fails at linking:

undefined reference to `etime_'

Is there a problem I cannot see, or is ETIME not yet included in gfortran ?

Thanks for you help,
FX

PS : please Cc me as I'm no suscriber of this mailing-list

[Tobias Schlüter]

Post by François-Xavier Coudert
Hello,
I am trying to use gfortran to compile scientific code (including BLAS,
LAPACK and ARPACK packages), but my compilation (using gfortran-20040922)
undefined reference to `etime_'
Is there a problem I cannot see, or is ETIME not yet included in gfortran ?

Hm, gfortran has both an etime subroutine and an etime function. I just tried
them, and they work (as documented in the g77 documentation, we don't yet seem
to have our own documentation). There's an issue that I found: if one tries to
use both the function and the subroutine in the same program unit, one gets a
compiler error.

Could you provide some example source code illustrating your problem?

- Tobi

[François-Xavier Coudert]

Post by Tobias Schlüter
Could you provide some example source code illustrating your problem?

Here is a minimal example, working well with g77.

pc31 ~ $ cat sec.f
PROGRAM MAIN

REAL SECOND
REAL T1
REAL TARRAY( 2 )
REAL ETIME
EXTERNAL ETIME

T1 = ETIME( TARRAY )
SECOND = TARRAY( 1 )

END

pc31 ~ $ g77 sec.f

pc31 ~ $ /opt/gfortran/gfortran-20040922/bin/gfortran sec.f
/tmp/ccypfi8q.o(.text+0xd): In function `MAIN__':
: undefined reference to `etime_'
collect2: ld returned 1 exit status

[Tobias Schlüter]

ETIME is declared EXTERNAL in LAPACK. g77 calls the intrinsic nonetheless, so

Post by François-Xavier Coudert
Here is a minimal example, working well with g77.
pc31 ~ $ cat sec.f
PROGRAM MAIN
REAL SECOND
REAL T1
REAL TARRAY( 2 )
REAL ETIME
EXTERNAL ETIME
T1 = ETIME( TARRAY )
SECOND = TARRAY( 1 )
END

We use different calling ocnventions for intrinsics and extrernal functions
(because we want to evade copying of arrays where possible), so I propose this
workaround.

I assume that g77 works with most of its intrinsics declared EXTERNAL, but
instead of trying to fix all of them, I suggest fixing them one at a time as
they appear. If we want to fix all at a time, we should have a field
indicating "this intrinsic takes its arguments in the g77 way", and not use
prefix on their implementations.

Anyway, patch below. Built and verified that it works as expected.

- Tobi

2004-09-24 Tobias Schlueter <***@physik.uni-muenchen.de>

* etime.c (etime_): New function.

Index: etime.c
===================================================================
RCS file: /cvs/gcc/gcc/libgfortran/intrinsics/etime.c,v
retrieving revision 1.2
diff -u -p -r1.2 etime.c
--- etime.c 21 Jun 2004 22:25:12 -0000 1.2
+++ etime.c 24 Sep 2004 12:07:38 -0000
@@ -78,3 +78,22 @@ prefix(etime) (gfc_array_r4 *t)
prefix(etime_sub) (t, &val);
return val;
}
+
+/* LAPACK's test programs declares ETIME external, therefore we
+ need this. */
+
+GFC_REAL_4
+etime_ (GFC_REAL_4 *t)
+{
+ gfc_array_r4 desc;
+ GFC_REAL_4 val;
+
+ /* We only fill in the fields that are used in etime_sub. */
+ desc.dim[0].lbound = 0;
+ desc.dim[0].ubound = 1;
+ desc.dim[0].stride = 1;
+ desc.data = t;
+
+ prefix(etime_sub) (&desc, &val);
+ return val;
+}

[Paul Brook]

Post by Tobias Schlüter
* etime.c (etime_): New function.

Ok.

Paul

[François-Xavier Coudert]

Post by Tobias Schlüter
Could you provide some example source code illustrating your problem?

OK, using the code I posted in the brother message of this one, I can get it
working if I remove the 'external ETIME' declaration.

I don't know if it's a normal situation (I suppose not), but at least I can
work around this issue.

Sorry for the annoyance,
FX