[OMPI users] errors in stdin
Hi, I am currently running the code by using mpirun -np 4 abc.exe < test.in>out abc.exe is my executed file, I list my input values in test.in as AAA value1 BBB value2 I run the same code by using intel mpi, it's fine, it can read parameters from stdin. But when I use open mpi, it's incorrect. It seems openmpi just put some strange character into the stdin, so my code dosen't accept that parameters and it stops. Dose anyone has this kind of problems. thanks brian
[OMPI users] mpi_rank : use as function arguments, or find it again every time
Hi, I need to use rank of the mpi processes in my sub functions ( c language). which one has more cost?; creating a new variable& and finding the rank in each sub function by: int mpi_rank; MPI_Comm_rank(MPI_COMM_WORLD,&mpi_rank); or finding the rank in main function once, and every time sending this information to thefunctions in function arguments like foo(a,b,c,mpi_rank);
Re: [OMPI users] AlphaServers & OpenMPI
Jeff Squyres wrote: > > This specific problem may be due to a bug that Brian just found/fixed > in the configure script last night (due to a bug report from Paul Van > Allsburg). Could you try any nightly trunk tarball after r14600 (the > fix hasn't made its way over to the 1.2 release branch yet; I assume > it will soon)? I got the nightly built tarball from version 1.3a1r14639, but now the make ends with: libtool: compile: /opt/gcc/bin/gcc -DHAVE_CONFIG_H -I. -I../../../../opal/include -I../../../../orte/include -I../../../../ompi/include -DPKGDATADIR=\"/opt/openmpi/share/openmpi\" -I../../../.. -O3 -DNDEBUG -mfp-trap-mode=su -finline-functions -fno-strict-aliasing -pthread -MT btl_openib_component.lo -MD -MP -MF .deps/btl_openib_component.Tpo -c btl_openib_component.c -fPIC -DPIC -o .libs/btl_openib_component.o btl_openib_component.c: In function 'btl_openib_async_thread': btl_openib_component.c:1203: error: 'IBV_EVENT_CLIENT_REREGISTER' undeclared (first use in this function) btl_openib_component.c:1203: error: (Each undeclared identifier is reported only once btl_openib_component.c:1203: error: for each function it appears in.) make[2]: *** [btl_openib_component.lo] Error 1 make[2]: Leaving directory `/home/Software/openmpi-1.3a1r14639/ompi/mca/btl/openib' make[1]: *** [all-recursive] Error 1 make[1]: Leaving directory `/home/Software/openmpi-1.3a1r14639/ompi' make: *** [all-recursive] Error 1 Here something is really broken in the code !?!? Rob. - Get the free Yahoo! toolbar and rest assured with the added security of spyware protection.
Re: [OMPI users] AlphaServers & OpenMPI
On May 12, 2007, at 8:25 AM, Rob wrote: > This specific problem may be due to a bug that Brian just found/ fixed > in the configure script last night (due to a bug report from Paul Van > Allsburg). Could you try any nightly trunk tarball after r14600 (the > fix hasn't made its way over to the 1.2 release branch yet; I assume > it will soon)? I got the nightly built tarball from version 1.3a1r14639, but now the make ends with: Woo hoo! You got significantly farther along than you did last time. libtool: compile: /opt/gcc/bin/gcc -DHAVE_CONFIG_H -I. - I../../../../opal/include -I../../../../orte/include -I../../../../ ompi/include -DPKGDATADIR=\"/opt/openmpi/share/openmpi\" - I../../../.. -O3 -DNDEBUG -mfp-trap-mode=su -finline-functions -fno- strict-aliasing -pthread -MT btl_openib_component.lo -MD -MP - MF .deps/btl_openib_component.Tpo -c btl_openib_component.c -fPIC - DPIC -o .libs/btl_openib_component.o btl_openib_component.c: In function 'btl_openib_async_thread': btl_openib_component.c:1203: error: 'IBV_EVENT_CLIENT_REREGISTER' undeclared (first use in this function) This stuff was just added the other day (remember: trunk = active development). I suspect that this is using an OpenFabrics interface that was not always available. What version of OFED do you have installed, 1.0? We can probably add a proper configure test for IBV_EVENT_CLIENT_REREGISTER and protect those sections of the code for OFED installations that do not have it. -- Jeff Squyres Cisco Systems
Re: [OMPI users] AlphaServers & OpenMPI
Jeff Squyres wrote: > > This stuff was just added the other day (remember: trunk = active > ... > > What version of OFED do you have installed, 1.0? We can probably add > a proper configure test for IBV_EVENT_CLIENT_REREGISTER and > protect those sections of the code for OFED installations that do not have it. OFED? Protect sections? No idea what you are talking about here. I checked the available packages for CentOS-alpha, but there's nothing which has "ofed" or "fabric" in its name. I myself never heard of such a thing, so I don't know what to do next. I even have no idea what you mean by "protect sections of the code". CentOS follows the RedHat enterprise editions closely. My version is 4.4, which is the latest for AlphaServer systems (CentOS 5 is recently released, but not for AlphaServers). So I wonder if this OFED thing is something too recent Rob. - Be a better Heartthrob. Get better relationship answers from someone who knows. Yahoo! Answers - Check it out.
Re: [OMPI users] AlphaServers & OpenMPI
On May 12, 2007, at 8:49 AM, Rob wrote: > What version of OFED do you have installed, 1.0? We can probably add > a proper configure test for IBV_EVENT_CLIENT_REREGISTER and > protect those sections of the code for OFED installations that do not have it. OFED? Protect sections? No idea what you are talking about here. I checked the available packages for CentOS-alpha, but there's nothing which has "ofed" or "fabric" in its name. I myself never heard of such a thing, so I don't know what to do next. OFED is the OpenFabrics Enterprise Distribution; the open source InfiniBand and iWARP network driver stack that is jointly developers by a consortium of vendors. The OFED software is included in several Linux distributions, although it tends to release a bit faster than the Linux distros release. For example, OFED is just about to release v1.2, although some distros are still shipping OFED v1.0. I don't know what version of OFED CentOS includes. I even have no idea what you mean by "protect sections of the code". I'm referring to the fact that we need to setup the code so that it can correctly compile even if you have an older version of the OFED libraries that do not provide the IBV_EVENT_CLIENT_REREGISTER functionality. If you do not have InfiniBand or iWARP hardware, you can disable this whole code path with: ./configure --without-openib ... CentOS follows the RedHat enterprise editions closely. My version is 4.4, which is the latest for AlphaServer systems (CentOS 5 is recently released, but not for AlphaServers). So I wonder if this OFED thing is something too recent Rob. Be a better Heartthrob. Get better relationship answers from someone who knows. Yahoo! Answers - Check it out. ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- Jeff Squyres Cisco Systems
Re: [OMPI users] mpi_rank : use as function arguments, or find it again every time
Hello, On Saturday 12 May 2007 11:25, Sefa Arslan wrote: > I need to use rank of the mpi processes in my sub functions ( c > language). which one has more cost?; > creating a new variable& and finding the rank in each sub function by: > int mpi_rank; > MPI_Comm_rank(MPI_COMM_WORLD,&mpi_rank); > or finding the rank in main function once, and every time sending > this information to thefunctions in function arguments like > foo(a,b,c,mpi_rank); Depends how often You call this function, too. But You ommitted another solution, so let's see: 1.) Calling MPI_Comm_rank involves: - Overhead of calling the function MPI_Comm_rank - Overhead of checking the communicator (in case of --enable-debug), - (otherwise, only the inline function ompi_comm_rank is evaluated as comm->c_my_rank). 2.) Calling MPI_Comm_rank once and pass the rank as parameter: - Clutters Your function parameters... 3.) Keep the mpi_rank of MPI_COMM_WORLD as global variable. The latter would be the fastest; although not possibly the nicest solution; if You already have a structure keeping init-once variables, this would be the place to keep... Hope, this helps. With best regards, Rainer -- Dipl.-Inf. Rainer Keller http://www.hlrs.de/people/keller High Performance Computing Tel: ++49 (0)711-685 6 5858 Center Stuttgart (HLRS) Fax: ++49 (0)711-685 6 5832 POSTAL:Nobelstrasse 19 email: kel...@hlrs.de ACTUAL:Allmandring 30, R.O.030AIM:rusraink 70550 Stuttgart
Re: [OMPI users] errors in stdin
This is unfortunately a known problem -- see: https://svn.open-mpi.org/trac/ompi/ticket/967 I mostly fixed the problem a few weeks ago, but was called away on other higher-priority items before fully finishing and testing the fix. :-( The workaround until this is fixed is to read from a file instead of stdin. On May 12, 2007, at 3:45 AM, anyi li wrote: Hi, I am currently running the code by using mpirun -np 4 abc.exe < test.in>out abc.exe is my executed file, I list my input values in test.in as AAA value1 BBB value2 I run the same code by using intel mpi, it's fine, it can read parameters from stdin. But when I use open mpi, it's incorrect. It seems openmpi just put some strange character into the stdin, so my code dosen't accept that parameters and it stops. Dose anyone has this kind of problems. thanks brian ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- Jeff Squyres Cisco Systems
Re: [OMPI users] AlphaServers & OpenMPI
Jeff Squyres wrote: > > If you do not have InfiniBand or iWARP hardware, you can disable this > whole code path with: > ./configure --without-openib ... BINGO! Both of my problems are gone with this configure flag and the make ends successfully, also with the --enable-shared configure flag. After installing (with the shared .so libraries), I also get working mpi-compilers. Note: when I do the above with the stable release 1.2.1, I get again the following when compiling code: $ mpicc mycode.c gcc: dummy: No such file or directory gcc: ranlib: No such file or directory But the good news is that the latter problem disappears with the nightly built openmpi-1.3a1r14639. Thus I'm looking forward to the next release that will support my AlphaServers out-of-the-box. Although I'm not at all an MPI expert, but if tests are needed to help openmpi work on AlphaServers, then I am willing to help with tests on my systems. By the way: does OpenMPI support running MPI on a variety of architectures? Can I run MPI code parallel on Intel-PCs, HP/Itaniums and AlphaServers (all running Linux/CentOS) ? Hence, is this kind of cross-architecture supported in OpenMPI? Regards, Rob. - Sick sense of humor? Visit Yahoo! TV's Comedy with an Edge to see what's on, when.
Re: [OMPI users] AlphaServers & OpenMPI
Hi, One more question with respect to running OpenMPI on my AlphaServers and other PCs in my lab. Version 1.2.1 fails to build on the AlphaServers, but miraculously the nightly snapshot openmpi-1.3a1r14639 works fine. Now, I need to have OpenMPI set up on a production cluster, but openmpi-1.3XXX is under active development. What is recommended to use in this case for my cluster system? I would prefer that the appropriate fix also be applied to 1.2.x series, so that I can rely on the a stable version ? Thanks, Rob. - Be a better Heartthrob. Get better relationship answers from someone who knows. Yahoo! Answers - Check it out.
