Newsgroups: comp.programming.threads,comp.parallel.mpi From: Christoph Giess Subject: Re: Threads, MPI and Re-entrancy Organization: DKFZ Date: Mon, 06 Jul 1998 07:59:48 +0200 Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <35A067D4.238B@DKFZ-Heidelberg.de> Girindra D. Sharma wrote: > > Hi, > Has anybody used pthreads and MPI together ? Yes. > I have written a threaded pingpong using > the MPICH implementation of MPI. > The main thread hands over the messages to a > sender thread and does the receiving itself. > The pingpong is not behaving correctly. > It hangs in some MPI calls. So the thread > making that call blocks. > I think I might be facing re-entrancy problems. That's correct. > I know that MPI is re-entrant. Unfortunately not. > But the calls MPI might be making may not be > re-entrant. So I tried locking all my MPI > calls. This time the pingpong worked fine. > Am I correct in inferring the the underlying > system Iam using is not re-entrant. Yes. > How can I confirm this ? Have a look at the source code :-). There must be a locking. To get a simple model, I assumed in my work that all MPI calls are protected by a mutex semaphore. > I don't want to sequentialize my MPI calls. > Is there any way out ? It depends on what you want to do, but generally: NO. > Iam using SUN OS 5.6 That should be the same on all platforms using p4. > > Thanks in advance. > > -- Girindra Sharma Bye, Christoph ... still running against the wind. ======================================================================== Christoph Giess Deutsches Krebsforschungszentrum (German Cancer Research Center) Div. Medical and Biological Informatics H0100 Tel: (+49) 6221-42 2325 Im Neuenheimer Feld 280 Fax: (+49) 6221-42 2345 D-69120 Heidelberg e-mail Ch.Giess@DKFZ-Heidelberg.de Germany http://mbi.dkfz-heidelberg.de/ ========================================================================