839
Comment:
|
1332
|
Deletions are marked like this. | Additions are marked like this. |
Line 4: | Line 4: |
=== Portland Compiler === |
|
Line 6: | Line 8: |
On the Ubuntu machines is the PGF compiler suite installed that has MPICH implemented. | On the Ubuntu machines the PGF compiler suite is installed that has MPICH implemented. |
Line 15: | Line 17: |
. /opt/pgi/linux86-64/8.0-2/mpi.sh | . /opt/pgi/linux86-64/10.4/mpi.sh |
Line 21: | Line 23: |
. /opt/pgi/linux86/8.0-2/mpi.sh | . /opt/pgi/linux86/10.4/mpi.sh |
Line 46: | Line 48: |
=== Intel Compiler === The MPICH2 Library (1.3.2p1) is installed in directory ''/usr/local/icg/icg1/local/mpich2-intel'' (for Intel C and Fortran compilers). Add the bin subdirectory of the installation directory to your path: {{{ PATH=/usr/local/icg/icg1/local/mpich2-intel/bin:$PATH export PATH }}} The programs can be compiled with: {{{ make useComp=ifc useMPI=true <progname> }}} Execute on <n> CPUs: {{{ mpirun -np <n> <progname> }}} |
HowToParallel
Portland Compiler
It is possible to run CLaMS on multi-processor machines using MPI or OpenMP.
On the Ubuntu machines the PGF compiler suite is installed that has MPICH implemented.
In the following a short How to:
First it is necessary to make the path of the MPI libraries public by executing:
for 64bit machines:
. /opt/pgi/linux86-64/10.4/mpi.sh
for 32bit machines:
. /opt/pgi/linux86/10.4/mpi.sh
Then compile the needed programs by executing:
make useMPI=true
Before running the CLaMS script or a single program you have to execute:
ssh-add
This is necessary, because the processes communicate per ssh.
The programs must be called as following:
mpirun -np # program # - number of processors
Intel Compiler
The MPICH2 Library (1.3.2p1) is installed in directory /usr/local/icg/icg1/local/mpich2-intel (for Intel C and Fortran compilers).
Add the bin subdirectory of the installation directory to your path:
PATH=/usr/local/icg/icg1/local/mpich2-intel/bin:$PATH export PATH
The programs can be compiled with:
make useComp=ifc useMPI=true <progname>
Execute on <n> CPUs:
mpirun -np <n> <progname>