[GRASS-dev] New grass library and modules

[Soeren Gebbert]

Hi Markus,
thanks for the compliment. :)

I hope this small library is the first step to
establish parallel computation in grass, besides the MPI modules.

But important to say: this library only works in parallel on SMP or ccNUMA 
machines. This little presentation i made will explain a bit of the technical 
background (sorry its only available in German):

http://www-pool.math.tu-berlin.de/~soeren/grass/modules/grass-pde-lib/OpenMP_GPDE_presenation_handouts.pdf

Best regards
Soeren


Am Mittwoch, 24. Januar 2007 15:59 schrieb Markus Neteler:
> Hi Soeren,
>
> this looks really nice and will help to better establish GRASS
> as GIS workhorse also for heavy computations.
>
> You could submit this to Computer & Geosciences as article :)
>
> Markus
>
> On Mon, Jan 22, 2007 at 10:10:19PM +0100, Soeren Gebbert wrote:
> > Dear developers,
> > i would like to submit a new library and two modules based on the new
> > library into cvs.
> >
> > The new library called "gpde_lib" is designed to support the solution of
> > partial differential equations within grass. The discretization of PDE's
> > is based on raster and volume maps.
> > I have implemented groundwater flow for two and three dimensions.
> > Heat flow and solute transport are on my todo list.
> >
> > All library functions start with a N_
> > eg: N_create_les() or N_alloc_array_2d().
> >
> > The current draft doxygen documentation is available here:
> >
> > http://www-pool.math.tu-berlin.de/~soeren/grass/modules/grass-pde-lib/htm
> >l/
> >
> > I have implemented two modules based on the new library to solve confined
> > and unconfined groundwater flow in two dimensions and confined
> > groundwater flow in three dimensions. The two programs are still very
> > basic and support only two boundary conditions (dirichlet and neumann).
> >
> > The modules are called r.gwflow and r3.gwflow.
> >
> > The grass PDE lib provied functions to create and solve lineare equation
> > systems. Therefor different linear equation solvers are implemented:
> > * LU decomposition for quadratic matrices (direct)
> > * GAUSS elimination for quadratic matrices (direct)
> > * CG method for quadratic and sparse matrices (iterative)
> > * BiCGStab method for quadratic and sparse matrices (iterative)
> >
> > The solvers and the creation of a linear equation system (matrix
> > assembling) are parallelized with OpenMP http://www.openmp.org. The
> > speedup of the solvers is quite good (IMHO):
> > http://www-pool.math.tu-berlin.de/~soeren/grass/modules/screenshots/sgi_a
> >ltix_cg_bench.png
> >
> > The new library do not use functions of the gmath library to create
> > vectors or matrices, or to solve a linear equation systems. There reasons
> > are: 1.) I was quite unsure if the gmath library calls are thread safe
> > 2.) I do not wanted to use function calls to get access to vectors and
> > matrices for performance reasons
> > 3.) I wanted to parallelize the equation solvers with OpenMP and therefor
> > all vector and matrix operations i use
> > 4.) I'm not very familiar with the gmath library
> > 5.) I do not know which gmath functions need LAPACK or BLAS
> >
> > If those question can be clarified, it should be possible to use gmath
> > functions in the gpde library.
> > The only problems i see for now are, that there are two different
> > function calls to perform a LU decomposition and the gpde lib do not use
> > the matrix and vector structures of the gmath lib.
> >
> > I have implemented a lot of unit and integration tests on code basis,
> > which will also be commited into cvs. Those tests assure the correct
> > functionality of the lib. If the tests are compiled, a new module called
> > test.gpde.lib is created and can be executed within grass to test all
> > functions of the library. IMHO we should create a test structure for each
> > library to check the correct functionality on function base.
> >
> > If there are no objections against the library, i would like to commit it
> > to cvs.
> >
> > Best regards
> > Soeren
> >
> > _______________________________________________
> > grass-dev mailing list
> > grass-dev at grass.itc.it
> > http://grass.itc.it/mailman/listinfo/grass-dev
>
> _______________________________________________
> grass-dev mailing list
> grass-dev at grass.itc.it
> http://grass.itc.it/mailman/listinfo/grass-dev