Re: [OMPI users] Making RPM from source that respects --prefix
On Oct 7, 2009, at 9:36 PM, Bill Johnstone wrote: Thank you for your responses. Based on the information you both provided, I was able to get buildrpm to make the OMPI RPM the way I wanted. I ended up having to define _prefix , _mandir , and _infodir . Ah, yes -- _mandir and _infodir. Sorry; I should have mentioned those, but I always forget that some distros don't automatically put those dirs under _prefix. Sigh. Additionally, I found I had to use --define "shell_scripts_basename mpivars" because without that, when I tried to use mpi-selector, mpi- selector did not find the installation since it specifically seems to look for the shell scripts as mpivars.{sh,csh} rather than mpivars-1.3.3.{sh,csh} as the .spec file builds. I think the .spec file should be changed to match what mpi-selector expects. The intent of putting the version string at the end of the shell script names was to allow them to co-exist with other Open MPI installs in the same tree. In hindsight, that seems a little silly, because the scripts are installed in bindir -- where you can clearly only have 1 Open MPI installed at a time. I'll fix. Jeff, it might also be really useful to have a .spec build option to allow the RPM to register itself as the system default. I hand- modified the .spec file to do this. Please let me know if I should register a feature request somewhere more formally. Sure, that sounds reasonable. What --define option did you add for it? Something like --define "mpi_selector_default 1" that causes -- system to be added to the mpi-selector command lines in %post and %preun? -- Jeff Squyres jsquy...@cisco.com
[OMPI users] Compilation problems of OpenMPI 1.3.3 with Snow Leopard
Following my previous post regarding the problems that i am having to compile OpenMPI 1.3.3 on Snow Leopard, here are some observations: - OpenMPI 1.2.9 compiles fine on Snow Leopard. Staring with OpenMPI 1.3, the compilation fails with the error message described in my previous post. Note that OpenMPI 1.3 was compiling fine on Mac OS 10.5 Leopard. - OpenMPI 1.3.3 compiled fine on a system running Mac OS 10.4.11 Tiger. Best regards, Hakime Seddik Glacier and Ice Sheet Research Group Tel:+81-(0)90-6263-4074 email: hak...@pop.lowtem.hokudai.ac.jp Institute of Low Temperature Science Hokkaido University, Kita-19, Nishi-8, Kita-ku Sapporo 060-0819. Japan
Re: [OMPI users] MPI_Comm_accept()/connect() errors
The PATH variable contains /home/hx019035/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/ bin:/usr/games:/usr/local/maui/bin/: /home/hx019035/bin contains the local installation of OMPI 1.3.3 The LD_LIBRARY_PATH variable contains /home/hx019035/lib: These variables are being set in the .profile file on the hpcc00 node. Would there be a change anywhere else? From: Ralph Castain [mailto:rhc.open...@gmail.com] On Behalf Of Ralph Castain Sent: 07 October 2009 13:32 To: Blesson Varghese Subject: Re: [OMPI users] MPI_Comm_accept()/connect() errors Yes, it does. But the error message indicates a 1.2 version is running on hpcc00. On Oct 7, 2009, at 5:46 AM, Blesson Varghese wrote: Just a quick question. Would mpirun -version give me the version of the mpirun being executed? I am getting the result of that as 1.3.3. From: Ralph Castain [mailto:rhc.open...@gmail.com] On Behalf Of Ralph Castain Sent: 07 October 2009 11:58 To: Blesson Varghese Subject: Re: [OMPI users] MPI_Comm_accept()/connect() errors Hate to tell you this, but your output clearly indicates you are NOT running 1.3.3 - that is an output from a 1.2.x version of OMPI. Check you path and ld_library_path - you're still picking up the 1.2.5 version somewhere. On Oct 7, 2009, at 4:05 AM, Blesson Varghese wrote: Hi, Please refer to the emails below. I have made an upgrade to Open MPI 1.3.3 as suggested. The necessary environment variables have all been set. Attaching the output of ompi_info -all. However, the errors continue to persist. [hpcc00:31864] [0,0,0] ORTE_ERROR_LOG: Not found in file dss/dss_unpack.c at line 209 [hpcc00:31864] [0,0,0] ORTE_ERROR_LOG: Not found in file communicator/comm_dyn.c at line 186 [hpcc00:31864] *** An error occurred in MPI_Comm_connect [hpcc00:31864] *** on communicator MPI_COMM_WORLD [hpcc00:31864] *** MPI_ERR_INTERN: internal error [hpcc00:31864] *** MPI_ERRORS_ARE_FATAL (goodbye) The server program is as follows: #include #include #include int main( int argc, char **argv ) { MPI_Comm client; MPI_Status status; char port_name[MPI_MAX_PORT_NAME]; int buf; int size, again; MPI_Info portInfo; MPI_Init( , ); MPI_Comm_size(MPI_COMM_WORLD, ); MPI_Open_port(MPI_INFO_NULL, port_name); printf("server available at %s\n",port_name); MPI_Comm_accept(port_name, MPI_INFO_NULL, 0, MPI_COMM_WORLD, ); MPI_Recv(, 1, MPI_INT, MPI_ANY_SOURCE, MPI_ANY_TAG, client, ); MPI_Comm_disconnect( ); } The client program is as follows: #include #include #include #include int main( int argc, char **argv ) { MPI_Comm server; int buf = 8; char port_name[MPI_MAX_PORT_NAME]; MPI_Info portInfo; MPI_Init( , ); strcpy(port_name, "0.0.0:2000"); //The port name is hardcoded since 0.0.0:2000 is generated by the server program MPI_Comm_connect(port_name, MPI_INFO_NULL, 0, MPI_COMM_WORLD, ); MPI_Send(, 1, MPI_INT, 0, 1, server ); MPI_Comm_disconnect( ); MPI_Finalize(); return 0; } Would you please advise? Regards, Blesson. -Original Message- From: Blesson Varghese [mailto:hx019...@reading.ac.uk] Sent: 03 October 2009 12:20 To: 'Jeff Squyres' Subject: RE: [OMPI users] MPI_Comm_accept()/connect() errors Thank you. I shall try the upgrade very soon. -Original Message- From: Jeff Squyres [mailto:jsquy...@cisco.com] Sent: 03 October 2009 12:18 To: Blesson Varghese Subject: Re: [OMPI users] MPI_Comm_accept()/connect() errors On Oct 3, 2009, at 7:14 AM, Blesson Varghese wrote: > Thanks for your reply Jeff. Since, it is a teaching cluster of the > University, I am quite unsure if I would be able to upgrade it very > soon. > > Do you reckon that the error is due to the Open MPI version? > You can always install your own version of Open MPI under your $HOME or somesuch -- there is no requirement that Open MPI is installed by root in a central location. That being said, you might want to check with your administrator to ensure that this is ok with local policies -- see if they did any special setup for Open MPI, etc. But yes, we made a bunch of COMM_SPAWN improvements since the 1.2 series. -- Jeff Squyres jsquy...@cisco.com From: Blesson Varghese [mailto:hx019...@reading.ac.uk] Sent: 01 October 2009 12:01 To: 'Open MPI Users'; 'Ralph Castain' Subject: RE: [OMPI users] MPI_Comm_accept()/connect() errors The following is the information regarding the error. I am running Open MPI 1.2.5 on Ubuntu 4.2.4, kernel version 2.6.24 I ran the server program as mpirun -np 1 server. This program gave me the output port as 0.1.0:2000. I used this port name value as the command line argument for the client program: mpirun -np 1 client 0.1.1:2000. - The output of the "ompi_info --all" is attached with the email - PATH
[OMPI users] Re : Re : Yet another stdin problem
Thanks a lot ! will try this solution. Best regards. Zellabia. S De : Roman CheplyakaÀ : Open MPI Users Envoyé le : Mer 7 Octobre 2009, 17 h 42 min 58 s Objet : Re: [OMPI users] Re : Yet another stdin problem As a slight modification, you can write a wrapper script #!/bin/sh my_exe < inputs.txt and pass it to mpirun. 2009/10/7 Kilou Zelabia : > Ok thanks! > That's a solution but i was wondering if there could exist a more elegant > one ? means without any modification at the source level > > > De : Roman Cheplyaka > À : Open MPI Users > Envoyé le : Mer 7 Octobre 2009, 17 h 06 min 55 s > Objet : Re: [OMPI users] Yet another stdin problem > > Why not modify your program to read inputs.txt instead of stdin? > > 2009/10/7 Kilou Zelabia : >> Dear all, >> >> I'm trying to lunch an MPI program using the command >> >> mpirun -np 64 my_exe < inputs.txt >> >> inputs.txt actually contains the entries that are read by all processes >> however it seems that openmpi redirect the stdin only to the processes >> with >> rank 0. with leads to a segmentation fault on the other processes. >> The option "-stdin rank" doesn't help since i need all processes read the >> inputs.txt file. >> >> I'v tested with openmpi 1.3 and 1.2 whitout success. >> >> Thanks in advance for you help. >> >> Zellabia. S >> >> >> >> >> ___ >> users mailing list >> us...@open-mpi.org >> http://www.open-mpi.org/mailman/listinfo.cgi/users >> > > > > -- > Roman I. Cheplyaka > > ___ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users > > ___ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users > -- Roman I. Cheplyaka ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users
[OMPI users] Compilation problems of OpenMPI 1.3.3 with Snow Leopard
Hi, I am having problems to compile the latest release of OpenMPI on Snow Leopard (Mac OS 10.6). I have installed the latest gfortran compiler (which can be found here: http://hpc.sourceforge.net/) and when trying to compile OpenMPI 1.3.3, i get some linking issues: Making all in tools/wrappers depbase=`echo opal_wrapper.o | sed 's|[^/]*$|.deps/&|;s|\.o$||'`;\ gcc "-DEXEEXT=\"\"" -I. -I../../../opal/include -I../../../orte/ include -I../../../ompi/include -I../../../opal/mca/paffinity/linux/ plpa/src/libplpa -I../../.. -D_REENTRANT -DNDEBUG -O5 -ftree- vectorize -finline-functions -fno-strict-aliasing -fvisibility=hidden -MT opal_wrapper.o -MD -MP -MF $depbase.Tpo -c -o opal_wrapper.o opal_wrapper.c &&\ mv -f $depbase.Tpo $depbase.Po cc1: note: -ftree-vectorize enables strict aliasing. -fno-strict- aliasing is ignored when Auto Vectorization is used. /bin/sh ../../../libtool --tag=CC --mode=link gcc -DNDEBUG -O5 - ftree-vectorize -finline-functions -fno-strict-aliasing - fvisibility=hidden -export-dynamic -o opal_wrapper opal_wrapper.o ../../../opal/libopen-pal.la -lutil libtool: link: gcc -DNDEBUG -O5 -ftree-vectorize -finline-functions - fno-strict-aliasing -fvisibility=hidden -o .libs/opal_wrapper opal_wrapper.o ../../../opal/.libs/libopen-pal.dylib -lutil Undefined symbols: "_opal_few", referenced from: _main in opal_wrapper.o "_opal_os_path", referenced from: _add_extra_includes in opal_wrapper.o _main in opal_wrapper.o "_opal_argv_insert", referenced from: _options_data_expand in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o "_opal_argv_count", referenced from: _add_extra_includes in opal_wrapper.o _options_data_expand in opal_wrapper.o _find_options_index in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o "_opal_basename", referenced from: _main in opal_wrapper.o "_opal_install_dirs_expand", referenced from: _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o "_opal_argv_append_nosize", referenced from: _add_extra_includes in opal_wrapper.o _main in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o "_opal_argv_split", referenced from: _add_extra_includes in opal_wrapper.o _options_data_expand in opal_wrapper.o _load_env_data_argv in opal_wrapper.o _main in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o "_opal_argv_delete", referenced from: _main in opal_wrapper.o _main in opal_wrapper.o "_opal_argv_free", referenced from: _options_data_expand in opal_wrapper.o _load_env_data_argv in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _data_callback in opal_wrapper.o _data_callback in opal_wrapper.o "_opal_argv_copy", referenced from: _main in opal_wrapper.o "_opal_finalize_util", referenced from: _main in opal_wrapper.o "_opal_show_help", referenced from: _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o "_opal_util_keyval_parse", referenced from: _main in opal_wrapper.o "_opal_install_dirs", referenced from: _main in opal_wrapper.o "_opal_init_util", referenced from: _main in opal_wrapper.o "_opal_path_findv", referenced from: _main in opal_wrapper.o "_opal_strerror", referenced from: _main in opal_wrapper.o "_opal_argv_join", referenced from: _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o _main in opal_wrapper.o ld: symbol(s) not found collect2: ld returned 1 exit status make[2]: *** [opal_wrapper] Error 1 make[1]: *** [all-recursive] Error 1 make: *** [all-recursive] Error 1 The same problem appears when trying to compile OpenMPI 1.3.2. Did someone have the same problem?