Differences between revisions 1 and 16 (spanning 15 versions)
⇤ ← Revision 1 as of 2009-12-10 13:30:11 →
Size: 799
Editor: RobertPommrich
Comment:
← Revision 16 as of 2016-08-30 08:37:37 → ⇥
Size: 1126
Editor: NicoleThomas
Comment:
Deletions are marked like this. Additions are marked like this.
Line 1: Line 1:
= HowToParallel = #acl ClamsUserGroup:read,write,delete,revert All:read
Line 3: Line 3:
It is possible to run CLaMS on multi-processor machines using MPI or OpenMP.
Line 5: Line 4:
On the Ubuntu machines is the PGF compiler suite installed that has MPICH implemented. ## page was renamed from HowToParallel
== ParallelHowTo ==
Line 7: Line 7:
In the following a short How to: <<TableOfContents>>
Line 9: Line 9:
First it is necessary to make the path of the MPI libraries public by executing: It is possible to run CLaMS on multi-processor machines using MPI.
Line 11: Line 11:
for 64bit machines:
Line 13: Line 12:
{{{
. /opt/pgi/linux86-64/8.0-2/mpi.sh		 === Intel Compiler 14.0.3 ===

 The MPICH Library is installed in directory ''/usr/nfs/local/local/mpich-intel''.
 Add the bin subdirectory of the installation directory of MPICH library to your path:
 {{{
PATH=/usr/nfs/local/local/mpich-intel/bin:$PATH
export PATH
Line 17: Line 21:
for 32bit machines:

{{{
. /opt/pgi/linux86/8.0-2/mpi.sh		  Remove serial compiled object files:
 {{{
make distclean
Line 23: Line 26:
Then compile the needed programs by executing:

{{{
make useMPI=true		  Compile with MPI:
{{{
make useMPI=true [useComp=ifc] progname
Line 29: Line 31:
Before running the CLaMS script or a single program you have to execute:

{{{
ssh-add		  Execute on <n> CPUs:
 {{{
mpiexec|mpirun -np <n> <progname>
Line 35: Line 36:
This is necessary, because the processes communicate per ssh. === Portland Compiler 14.6 ===
Line 37: Line 38:
The programs must be called as following:  It is necessary to make the path of the MPI libraries (included in the portland compiler suite) public by executing:
 {{{
. /opt/pgi/linux86-64/14.6/mpi.sh
}}}
Line 39: Line 43:
{{{
mpirun -np # program		  Remove serial compiled object files:
 {{{
make distclean
}}}
Line 42: Line 48:
# - number of processors  Compile with MPI:
 {{{
make useMPI=true [useComp=pgi] progname
}}}
Line 44: Line 53:
 Execute on <n> CPUs:
 {{{
mpiexec|mpirun -np <n> <progname>

ParallelHowTo

Contents

  1. ParallelHowTo
    1. Intel Compiler 14.0.3
    2. Portland Compiler 14.6

It is possible to run CLaMS on multi-processor machines using MPI.

Intel Compiler 14.0.3

  • The MPICH Library is installed in directory /usr/nfs/local/local/mpich-intel. Add the bin subdirectory of the installation directory of MPICH library to your path: 

    PATH=/usr/nfs/local/local/mpich-intel/bin:$PATH
export PATH
    Remove serial compiled object files: 
    make distclean
    Compile with MPI: 
    make useMPI=true [useComp=ifc] progname

    Execute on <n> CPUs: 

    mpiexec|mpirun -np <n> <progname>

Portland Compiler 14.6

  • It is necessary to make the path of the MPI libraries (included in the portland compiler suite) public by executing: 
    . /opt/pgi/linux86-64/14.6/mpi.sh
    Remove serial compiled object files: 
    make distclean
    Compile with MPI: 
    make useMPI=true [useComp=pgi] progname

    Execute on <n> CPUs: 

    mpiexec|mpirun -np <n> <progname>