Versatile Advection Code

Principal Contact Person and Organization (including e-mail address):

Gabor Toth, Eotvos University, Budapest, Hungary 
(gtoth@hermes.elte.hu) 
Rony Keppens, FOM Institue for Plasma Physics, Rijnhuizen, Holland
(keppens@rijnh.nl)

Brief Description of Application:

Hydrodynamic and magnetohydrodynamic problems in astrophysics. The code can handle 1, 2 and 3D problems as well.

Number of Lines of Code: 15000 (the HPF compatible part)

Target Platforms and HPF Compilers Used:

Any UNIX machine with a Fortran 77, Fortran 90 or HPF compiler (for parallel execution). Tested machines include: Linux PC, IBM, SUN, DEC, HP, SGI workstations, Cray C90, J90 vector computers, IBM SP, Cray T3E, T3D parallel machines. We used several

Coding Styles (data decompositions, computational methods):

Structured finite volume grid. Fully data parallel for the computational grid (array syntax, FORALL, WHERE). High resolution shock capturing methods (FCT, TVD).

Extrinsic Interfaces Used (and reasons):

None at the moment. We use BLAS libraries, but that part of the code is not HPF compatible yet. We plan to make it compatible by using HPF_LOCAL.

Performance Information, if Available (including any possible comparisons to MPI and/or OpenMP):

Parallel scaling as well as single node performance is satisfactory
     for the explicit time integration schemes.
     This means that for a big enough application the scaling is almost linear
     up to 10-20 processors, and the single node performance of the HPF compiler
     is comparable with that of the F90/F77 compiler.
     We do not have access to much bigger machines.
     Performance of VAC is discussed in detail in a conference proceedings:

     "Comparison of Different Computer Platforms for
      Running the Versatile Advection Code"
     by G. Toth and R. Keppens, 1998,
     Lecture Notes in Computer Science, 1241, 368-376
     (also available from http://hermes.elte.hu/~gtoth/Papers/vac.html)

Please comment on any aspects of the application that might be interesting, including any problems using HPF effectively:

Complicated CASE and IF statements also confused early versions of compilers. We also use namelists, and most HPF compilers still have bugs related to that. Our first experience with using subroutines locally failed due to the incorrect passing of array segments to the local subrouines. We know now that this is a limitation of HPF.

URL

http://www.phys.uu.nl/~toth