CFD Online Discussion Forums - autoparallelization

Dear Mr. Kayakol I am also a WinNT (friend, and) user and I have a CFD code called NAS3D (using the LES concept). I am able to run this code both on Windows (I have a graphical interface using SDK, but also I can run in a command line mode) or on any other platform (Unix) by only recompiling the source. I was looking for the same kind of tools (the autoparallelization needs a compiler to do so) as you do now, and I could not found any (for Win32 platforms). I had also the opportunity to test the auto parallelization facilities that the SGI compilers have now. While these auto parallelization may work for some class of algorithms (and these compilers are doing a very good job, I think the best in these area), in my case (Navier Stokes solver, using the Multidimensional Upwind approach) it proved him self totally unefficient. But, it was not too hard for me to parallelize my code using PVM (and I intend to have also the possibility to run MPI) and to do tests, on my WinNT computer. You can find and install (It has a very nice install shield facility, for Win32) the PVM 3.4beta 6, from: http://www.netlib.org/pvm3/win32/ !! It works excellent! From my experience, you can not expect too much speedup if you only intend to use the auto parallelization facility of a compiler. That depends on the problem that you have to solve. But excepting a limited number of cases, this option will give you always less performance than the one that you can obtain by using the Message Passing procedure. Again, is not too hard to do so, and it will give you huge satisfactions. Succes!

Best regards,

Doru Caraeni Aerospace Engineer, LTH , Sweden, Email: dc@mail.vok.lth.se phone:(046)-046-2224744