A little more on what was suggested by Michael,
We usually want our "mpi compiler wrapper" to use ifort (Intel Fortran)
opposed to gfortran (GNU Fortran). We also usually want to use LAM/MPI and
so I know it works for this, I guess however it would be possible to use the
following method for MPICH etc.
Method:
1) Install the OSCAR cluster on a clean head node machine as you normally
would.
2) Install Intel Fortran or whatever it is you want for your compiler(s).
3) Download LAM/MPI source code that matches the installed rpm version.
4) Compile the LAM/MPI source paying attention to the following configure
time switches:
FC=ifort F77=ifort FFLAGS=-ipo_obj LDFLAGS=-static-libcxa
--prefix=/opt/lam_mpi_ifort
Obviously FC and F77 set ifort as your Fortran compiler and the prefix means
you wont write over your existing lam/mpi bindings.
IMPORTANT: you will need to compile and link with all future code with
static libcxa else you have to push this library to all the nodes cexec cp
... ... There is little speed difference between a static and linked libcxa
(that I have noticed) as the likelihood of it being loaded in memory is very
small. The built binaries will obviously be larger as the library is
compiled into the code where needed.
TAKE NOTE: FFLAGS=-ipo_obj is there to force the ifort compiler to create
object files with the .o extension so that they can be found by the open
source linkers that will be expecting them.
NOTE: You are likely to get complaints from lots of things about missing
mpiof.h if you use MPI IO routines via ROMIO, therefore remember to use
-I/path_to_mpiof.h/ etc.
ALSO, a fortran compiler isn't actually required to compile lam/mpi "it only
uses the fortran compiler in the configure script to determine the symbol
style, datatype sizes, etc.", to quote Jeff Squyres
http://www.lam-mpi.org/MailArchives/lam/2003/06/6139.php, one of the lam/mpi
developers.
5) Create a script that can switch the pointing of your mpif77, mpif90, etc
that OSCAR installed in /opt to your ifort versions of those wrappers and
back again (or permanently link them if you prefer).
I also guess that you can do this more sensibly by getting the SRPM of the
OSCAR mpi version you want to have installed and edit the spec file with the
configure options as suggested above and build the rpm. Then use the built
rpm in place of the OSCAR mpi rpm of the same name whilst installing.
So there you are, how to get some MPI bindings to work with ifort. I guess
then you can compile whatever it is you want to compile happily with any
luck :)
Sorry for the long post, hope this helps....
Regards,
Stuart Archibald (Imperial College London)
On 10/09/2007, Michael Edwards <[EMAIL PROTECTED]> wrote:
>
> With some googling I found these notes on MCNPx and MPI
>
> http://mcnpx.lanl.gov/opendocs/installation/MPInotes
>
> I am guessing that our version of mpich was not compiled with the f90
> bindings as suggested in that document. You could download the source
> packages for OSCAR and rebuild mpich with the f90 bindings you need.
>
> I am not very familiar with what bindings OSCAR uses though, so I may be
> wrong.
>
> On 9/10/07, Annette Sahores <[EMAIL PROTECTED]> wrote:
> > I have installed OSCAR on RHEL4 with the Intel fortran compiler
> > 10.0.026.
> > When I try to compile MCNP5 (Monte Carlo) using mpif90, That comes with
> > OSCAR Y get the message:
> >
> > No Fortran 90 Compiler specified when mpif90 was created, or
> > configuration file does not specify a compiler.
> >
> > The configuration file points to:
> > MPIFC=/opt/mpich-ch_p4-gcc-1.2.7/bin/mpif90
> > The intel fortran compiler was installed before OSCAR, How I tell OSCAR
> > to use ifort when installs mpich?
> > Or how I tell mpich to use ifort as its fortran compiler?
> >
> > Thanks
> > --
> > Annette Sahores
> > INVAP S.E.
> > Centro de Cómputos y comunicaciones.
> > e-mail: [EMAIL PROTECTED]
> >
> >
> >
> -------------------------------------------------------------------------
> > This SF.net email is sponsored by: Microsoft
> > Defy all challenges. Microsoft(R) Visual Studio 2005.
> > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> > _______________________________________________
> > Oscar-devel mailing list
> > [email protected]
> > https://lists.sourceforge.net/lists/listinfo/oscar-devel
> >
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2005.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Oscar-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/oscar-devel
>
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Oscar-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/oscar-devel
