pawsen / SEM

Spectral methods for FEM

SEM

Spectral methods for FEM

Install and compile

Short: sudo apt-get install libvtk5-dev libblas-dev mpic++

vtk

Install headers and libraries for C++: sudo apt-get install libvtk5-dev

For a list of related packages: https://launchpad.net/ubuntu/+source/vtk

cblas

Documentation: http://www.netlib.org/lapack/explore-html/ or old-fashion http://www.netlib.org/blas/

Beside being implementations, BLAS and LAPACK are also API standard for basic linear algebra operations (such as vector and matrix multiplication). So one can test with calls to blas, and then change to a optimized implementation, eg. Atlas or Openblas later.

cblas is a part of the blas installation(actually it is just an interface for calling the fortran library blas.so). Pass all arguments by reference because of the Fortran implementation.

http://packages.ubuntu.com/source/quantal/blas sudo apt-get install libblas-dev

lapack

User guide http://www.netlib.org/lapack/lug/lapack_lug.html

http://www.math.utah.edu/software/lapack.html http://www.netlib.org/clapack/readme

Call fortran files from C/C++

http://www.math.utah.edu/software/c-with-fortran.html

Debugging MPI

Start by looking at http://stackoverflow.com/questions/329259/how-do-i-debug-an-mpi-program

And then at http://www.open-mpi.org/faq/?category=debugging

It is possible to use valgrind with MPI. There is two directions to go:

Compile OpenMPI linking to any Valgrind installation

OpenMPI have a memchecker included. It is enabled like this:

First MPI needs to be compiled with support for valgrind. Download Open MPI and configure as described here.

Then it is possible to run $ mpirun -np 8 valgrind ./my-exe

that is, launch 8 valgrinds, each running one process’ worth of my-exe

Compile valgrind linking to any OpenMPI installation

Valgrind needs to be compiled with support for mpi, as described here.

$./configure –with-mpicc=/path/to/mpicc

This will make a library file: libmpiwrap-<platform>.so(libmpiwrap-amd64-linux.so) which is needed when valgrind is called. Alternatively the lib can be linked directly to my-exe at compilation time. This should not influence the generel behavior of my-exe.

The finer details - difference between the methods

Look here for a technical description of the difference of the two methods.

Write .vtu file to paraview

http://packages.ubuntu.com/source/quantal/vtk

MPI: http://liberlocus.blogspot.dk/2013/03/how-to-write-vtk-file-in-parallel.html

http://vtk.org/gitweb?p=VTK.git;a=blob;f=Examples/DataManipulation/Cxx/RGrid.cxx

!HUSK! At binær data skal skrives som Big Indian. Standard er litle indian. Kun relevant hvis vi selv skriver rutinen.

Compile on gbar

Set the following.

module load python/2.7.3 python=’pyhton2.7’ module load VTK/5.10.1

Path to libraries on gbar ########

######## LIBBLAS = -L/appl/htopopt/RH61/atlas-3.9.49/lib/ -lcblas -latlas INCBLAS = -I/appl/htopopt/RH61/atlas-3.9.49/include/

#######

####### LIBVTK = -L/appl/VTK/5.10.1/lib/vtk-5.10/ -lvtkCommon -lvtkIO -lvtkFiltering INCVTK = -I/appl/VTK/5.10.1/include/vtk-5.10/

For profiling

module load: studio/12u3b mpi/studio