From: Rainer M. <ra...@tb...> - 2007-03-21 17:28:02
|
Dear Eric, I have forwarded your email to the developer email list, as others might be interested or have ideas on your remaining AIX problems. > very strangely, static int memoryExhaustion = 0; in solverError.c is not > initialized on AIX (on Linux it works fine), mhm... strange. Can you (or anyone else) think of a workaround to initialize this global variable to 0? We don't get any warning about this, so it seems to be ISO-C90/ANSI-C compatible. > Also, I found that only the serial versions of Sundials libraries are > used. My aim is now to convert cvodeSolver.c and odeModel.c to use > nvector_parallel.h I also haven't looked at the details of the parallel solvers in SUNDIALS. However, when looking at the examples given by SUNDIALS I remember that I had the impression that it might require to somehow parallelize the ODE system itself! This might be very difficult with our approach of arbitraty ODE models. But maybe I am wrong. It would be great if you could have a look at that! Generally, the SUNDIALS email list is very responsive and helpful. But we should try to find out what we can before contacting them. >if > that is successful, I'll happily share the adaptation with you. That would of course be a great contribution! BTW, we also need to fully implement the online TCC-based compilation of all right-hand-side functions. Andrew had implemented that on Windows and it didn't work out-of-the-box on our Linux systems here. We never had a look at it, but I think it will be probably be rather easy to implement it generally, based on optional SOSlib compilation linking to the TCC library. Rainer Rainer Machne wrote: > > -------- Original Message -------- > Subject: SOSlib progress on AIX / parallel support? > Date: Wed, 21 Mar 2007 16:28:11 -0000 > From: Eric Fernandez <efe...@ph...> > To: Rainer Machne <ra...@tb...> > > Dear Rainer, > > I hope you are well and that you enjoyed your holidays. > > I contacted you a couple of weeks ago about SOSlib that I wanted to > install on an AIX server. I finally managed to do it, thanks to a > debugging session. For some reason, the conversion of models from Level > 1 to Level 2 does not work, because of some issue with libsbml. Also, > very strangely, static int memoryExhaustion = 0; in solverError.c is not > initialized on AIX (on Linux it works fine), although compiled with gcc > 3.3.2, and of course leaded to the software stopping. Maybe a bug in > gcc-3.3.2 ? I had to deactivate the memory checks (I know, it's bad) > meanwhile. > > Also, I found that only the serial versions of Sundials libraries are > used. My aim is now to convert cvodeSolver.c and odeModel.c to use > nvector_parallel.h instead (and doing the adjustments for parallel > processing). I was wondering, have you already done this ? According to > my understanding of SOSlib, this should be possible and should not > interfere with events, as events are tested and triggered after each ODE > solving time step. However I admit I walk in the dark there. Anyway, if > that is successful, I'll happily share the adaptation with you. > > Kind regards, > > Dr Eric Fernandez > Senior Simulation Scientist > Physiomics plc > The Magdalen Centre > The Oxford Science Park > Oxford > OX4 4GA > UK > Tel: +44(0)1865 784986 > > > ------------------------------------------------------------------------- > Take Surveys. Earn Cash. Influence the Future of IT > Join SourceForge.net's Techsay panel and you'll get the chance to share your > opinions on IT & business topics through brief surveys-and earn cash > http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV > _______________________________________________ > sbmlsolver-devel mailing list > sbm...@li... > https://lists.sourceforge.net/lists/listinfo/sbmlsolver-devel |