Thank you all, my problem is solved!
I configured with --with-fc=/opt/intel/compiler70/ia32/bin/ifc, but
that still didn't solve my problem.
So I used "mpif77 dlcca_mpi.f90 -showme" as I have been told. Guess
what, the thing used the wrong compiler, it called f77 (which is
seemingly another for the g77).
So I deleted the whole lam-stuff, unpacked it once more, configured it
as needed and it worked.
So my mistake was that I first used configure with
--with-hf77=/opt/intel/compiler70/ia32/bin/ifc which was wrong (and
called make and make install), than I simply used configure with the
right parameter and installed it, but without doing make clean or
something similar or deleting the whole thing. End of my little
successtory.
Bye,
Nils
Am Montag, 02.06.03, um 19:33 Uhr (US/Pacific) schrieb Jeff Squyres:
> On Mon, 2 Jun 2003 Nils.Stritzel_at_[hidden] wrote:
>
>> I am using lam-mpi on linuxcluster running Suse 7.2. Recently I
>> updated
>> the Intel Fortran Compiler from version 5 to version 7, which broke my
>> lam-mpi (version 6.5.6) installation. That was why I updated lam to
>> the
>> most recent version 6.5.9.
>
> Did you see today's post from Konrad Karczewski?
>
> http://www.lam-mpi.org/MailArchives/lam/msg06058.php
>
>> I built the whole thing from source using:
>> ./configure --prefix=/opt/lam
>> --with-hf77=/opt/intel/compiler70/ia32/bin/ifc
>
> Note that this is not quite right. I think you meant
> "--with-fc=/opt/intel/compiler70/ia32/bin/ifc". You can also specify
> --with-rsh="ssh -x".
>
>> make
>> make install
>>
>> furthermore I set the following environment variable in a script :
>> export LAMRSH="ssh -x"
>
> If you use the --with-rsh option, above, you won't need to set LAMRSH.
>
>> export LAMHOME="/opt/lam"
>
> There is no need to set LAMHOME here, especially since you're not
> using it
> consistently for PATH and MANPATH, below.
>
>> export PATH="$LAMHOME/bin:$PATH"
>> export MANPATH="$MANPATH:/opt/lam/man"
>
> My recommendation would be to just set PATH and MANPATH directly:
>
> export PATH="/opt/lam/bin:$PATH"
> export MANPATH="$MANPATH:/opt/lam/man"
>
> That should be all you need.
>
>> When try to call it with a .f90 file I get :
>
> Also note that the name "hf77" is now deprecated. The name "mpif77" is
> preferred (although they are both synonomous, mpif77 is more
> descriptive
> of what it is doing).
>
>> hf77 dlcca_mpi.f90
>> /usr/i486-suse-linux/bin/ld:dlcca_mpi.f90: file format not recognized;
>> treating
>> as linker script
>> /usr/i486-suse-linux/bin/ld:dlcca_mpi.f90:1: parse error
>> collect2: ld returned 1 exit status
>>
>> with f77 file I get:
>> hf77 code125.f -o try_mpi
>> /usr/lib/gcc-lib/i486-suse-linux/2.95.3/libg2c.a(open.o): In function
>> `f_open': open.o(.text+0x4da): the use of `tempnam' is dangerous,
>> better use
>> `mkstemp'
>>
>> which works despite the warning.
>>
>> So now mpi won't accept f90 files and the compiler considers a file
>> ending in .f a Fortran 77 files and won't compile it.
>
> This unfortunately sounds like an Intel compiler issue, not a LAM/MPI
> issue. Remember that mpif77 (hf77) simply add a few flags to the
> command
> line and then turn around and invoke the back-end compiler (e.g., -I,
> -L,
> etc.). You can use the "-showme" flag to show that mpif77 would have
> invoked. For example:
>
> mpif77 dlcca_mpi.f90 -showme
>
> You might want to try writing a simple non-MPI fortran 90 compiler and
> try
> invoking the intel compiler against it and see if it compiles (I'm not
> too
> familiar with the intel compilers to know if they support fortran 90 or
> not, or if you need to use an alternate invocation name to get the f90
> compiler).
>
> --
> {+} Jeff Squyres
> {+} jsquyres_at_[hidden]
> {+} http://www.lam-mpi.org/
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
|
