Menu
▾
▴
Re: [Fink-beginners] OpenMPI conflicts: mpirun and om-mpirun
|
From: Todd K. <bo...@ma...> - 2009-08-19 00:22:25
|
Sorry, I should've replied to all. Please see the response below. Thanks, Todd On Tue, Aug 18, 2009 at 6:24 PM, Todd Krause <bo...@ma...>wrote: > Hi Jack, > > Thanks for the quick reply. I went through the procedure you mentioned > (adding a "fink selfupdate" between the two steps to make sure fink didn't > try to install openmpi *again* --- I missed that the first time, and fink > aborted). Unfortunately I still have the same error when I try to use make: > > 06:09 PM btpro:Gadget2> make -f > ../bobtodd_tests/milos_test01/makefiles/makefile.milos01 > > om-mpicc -O3 -Wall -DUNEQUALSOFTENINGS -DPEANOHILBERT -DWALLCLOCK > -DSYNCHRONIZATION -DISOTHERM_EQS -I/sw/include -I/sw/include -I/sw/include > -c -o main.o main.c > Cannot open configuration file > /sw/share/doc/openmpi/openmpi/om-mpicc-wrapper-data.txt > Error parsing data file om-mpicc: Not found > make: *** [main.o] Error 243 > 06:10 PM btpro:Gadget2> which om-mpicc > /sw/bin/om-mpicc > > To give you an idea of what the makefile contains, the relevant section is > > ifeq ($(SYSTYPE),"bobtoddpro") > CC = om-mpicc > OPTIMIZE = -O3 -Wall > GSL_INCL = -I/sw/include > GSL_LIBS = -L/sw/lib > FFTW_INCL= -I/sw/include > FFTW_LIBS= -L/sw/lib > #MPICHLIB = -L/usr/local/mpich2/mpich2-1.0.5p4/install/lib -lmpich > #MPICHINCL = -I/usr/local/mpich2/mpich2-1.0.5p4/install/include > HDF5INCL = -I/sw/include > HDF5LIB = -L/sw/lib -lhdf5 -lz -Wl#,"-R /sw/lib" > endif > > (this worked fine before, when I just had mpicc... it picked up openmpi's > commands fine; but I understand you renamed them to coexist with lam-mpi... > is there really a reason to keep lam-mpi around, since they stopped > development and moved to openmpi? Just asking...). The problem seems to be > that the system is looking for a file > "/sw/share/doc/openmpi/openmpi/xxxxxx-wrapper-data.txt", where "xxxxxx" > should be the same as what I've set CC to; but it finds nothing with > "om-mpicc-...". Here's the contents of that directory on my system: > > 06:10 PM btpro:Gadget2> ls /sw/share/doc/openmpi/openmpi/ > amca-param-sets/ help-orte-iof.txt > help-coll-sync.txt help-orte-odls-base.txt > help-dash-host.txt help-orte-ps.txt > help-ess-base.txt help-orte-rmaps-base.txt > help-hostfile.txt help-orte-rmaps-rr.txt > help-mca-base.txt help-orte-rmaps-seq.txt > help-mca-bml-r2.txt help-orte-runtime.txt > help-mca-coll-base.txt help-orte-snapc-base.txt > help-mca-param.txt help-orted.txt > help-mpi-api.txt help-orterun.txt > help-mpi-btl-base.txt help-plm-base.txt > help-mpi-errors.txt help-plm-rsh.txt > help-mpi-runtime.txt help-plm-slurm.txt > help-mpool-base.txt help-ras-base.txt > help-odls-default.txt help-ras-slurm.txt > help-ompi-crcp-base.txt help-rmaps_rank_file.txt > help-ompi-dpm-base.txt mpic++-vt-wrapper-data.txt > help-ompi-dpm-orte.txt mpic++-wrapper-data.txt > help-ompi-pubsub-orte.txt mpicc-vt-wrapper-data.txt > help-ompi-server.txt mpicc-wrapper-data.txt > help-ompi_info.txt mpicxx-vt-wrapper-data.txt@ > help-opal-carto-file.txt mpicxx-wrapper-data.txt@ > help-opal-crs-base.txt mpif77-vt-wrapper-data.txt > help-opal-crs-none.txt mpif77-wrapper-data.txt > help-opal-runtime.txt mpif90-vt-wrapper-data.txt > help-opal-util.txt mpif90-wrapper-data.txt > help-opal-wrapper.txt opalc++-wrapper-data.txt > help-orte-clean.txt opalcc-wrapper-data.txt > help-orte-filem-base.txt ortec++-wrapper-data.txt > help-orte-filem-rsh.txt ortecc-wrapper-data.txt > 06:12 PM btpro:Gadget2> > > From my current understanding, it seems that these filenames would be fine > if "om-mpicc" were still named "mpicc"; but since the change, the filenames > don't match what the wrapper's expecting. > > So does that mean I just have to go into that directory and change all the > filenames by hand? As a general policy, I try not to change things in /sw > by hand if I don't have to, for fear of messing things up.... > > Thanks, > Todd > > > On Tue, Aug 18, 2009 at 4:53 PM, Jack Howarth <ho...@br...>wrote: > >> Doh. I misread your message. This is a glitch from earlier attempts >> at supporting openmpi on case-sensitive filesystems. There were some >> problems with the logic which caused the previous openmpi package to >> nuke the files in the newly installed one. These issues are all >> resolved in the current openmpi packaging and all you need to do is >> execute... >> >> fink reinstall openmpi >> fink reinstall openmpi-dev >> >> to restore the missing files. >> Jack >> ps I didn't realize at the time that the PostRmScript of the previously >> installed package would be executing after the installation of the >> new upgrade package. I assumed that it would be executing only after >> the original packages files were removed but not before the new packages >> files were installed. Or is that a bug in fink? >> >> On Tue, Aug 18, 2009 at 02:58:31PM -0400, Alexander Hansen wrote: >> > -----BEGIN PGP SIGNED MESSAGE----- >> > Hash: SHA1 >> > >> > Todd Krause wrote: >> > > Hi, >> > > >> > > I recently installed the fftw-mpi package, which in turn required >> > > installation of the openmpi package (fine in my book). After finally >> > > realizing that fink named the new openmpi commands in /sw/bin to have >> > > "om-" prefixes, I ran make to compile the code Gadget-2, and received >> > > the following error: >> > > >> > > 01:03 PM btpro:Gadget2> make -f >> > > ../bobtodd_tests/milos_test01/makefiles/makefile.milos01 >> > > om-mpicc -O3 -Wall -DUNEQUALSOFTENINGS -DPEANOHILBERT -DWALLCLOCK >> > > -DSYNCHRONIZATION -DISOTHERM_EQS -I/sw/include -I/sw/include >> > > -I/sw/include -c -o main.o main.c >> > > Cannot open configuration file >> > > /sw/share/doc/openmpi/openmpi/om-mpicc-wrapper-data.txt >> > > Error parsing data file om-mpicc: Not found >> > > make: *** [main.o] Error 243 >> > > 01:03 PM btpro:Gadget2> which om-mpicc >> > > /sw/bin/om-mpicc >> > > >> > > Interestingly, the last time I made Gadget-2, I didn't need to use >> > > "om-mpicc"; simply "mpicc" seemed to find my /sw/bin/mpicc (which was >> > > openmpi) without problems. Since then, unfortunately, I had to remove >> > > fink and start from scratch, and this time I realized everything's >> > > called "om-xxxxx". I found a similar error in an earlier exchange on >> > > the fink listserve: >> > > >> > > >> http://www.opensubscriber.com/message/fin...@li.../7238283.html >> > > >> > > the relevant part of which is the following >> > > >> > > -------------- begin quoted text --------------- >> > > The first problem was a file which the executable needed, that had an >> > > unexpected name: >> > > >> > > ---- >> > > >> > > nell:~/tmp/tst barnes$ om-mpif77 >> > > Cannot open configuration file >> > > /sw/share/doc/openmpi/openmpi/om-mpif77-wrapper-data.txt >> > > Error parsing data file om-mpif77: Not found >> > > >> > > nell:~/tmp/tst barnes$ sudo cp >> > > /sw/share/doc/openmpi/openmpi/mpif77-wrapper-data.txt >> > > /sw/share/doc/openmpi/openmpi/om-mpif77-wrapper-data.txt >> > > >> > > nell:~/tmp/tst barnes$ ls -l /sw/share/doc/openmpi/openmpi/*77* >> > > -rw-r--r-- 1 root admin 668 Jul 26 18:42 >> > > /sw/share/doc/openmpi/openmpi/mpif77-wrapper-data.txt >> > > -rw-r--r-- 1 root admin 668 Jul 26 18:49 >> > > /sw/share/doc/openmpi/openmpi/om-mpif77-wrapper-data.txt >> > > >> > > nell:~/tmp/tst barnes$ om-mpif77 >> > > gfortran: no input files >> > > >> > > ----- >> > > There were many other similar .txt files which probably also need >> > > renaming. >> > > So that's fixed with a kludge, but now it's having trouble finding >> > > libraries >> > > which do exist (perhaps the same .so vs .dynlib problem?): >> > > >> > > /usr/bin/ld: can't locate file for: -lmpi_f77 >> > > collect2: ld returned 1 exit status >> > > make: *** [table.x] Error 1 >> > > >> > > -------------- end quoted text ---------------- >> > > >> > > So my question is this: I can find the relevant data files, and I >> could >> > > rename them all to have the "om-" prefix, but is there a better fix so >> > > that I don't have to go through all that and muck around in /sw? >> > > Moreover, it seems like I'll have to do that for each wrapper I use >> > > (om-mpic++ next, then om-mpif77 if I have to use Fortran, etc.). And >> in >> > > the end, even that will only get me to the library error, according to >> > > the above email. Is there a resolution to this problem that I failed >> to >> > > find? >> > > >> > > Thanks very much, >> > > Todd >> > > >> > > >> > >> > I'll cc the maintainer for the first part. >> > >> > As for the the library linker issue above, it's hard to say without >> > seeing what is being fed to ld. However, it may be as simple as not >> > having '-L/sw/lib/openmpi' before '-lmpi_f77', since libmpi_f77.dylib is >> > not in ld's default search path. How to fix this depends on what the >> > package you're building supports. >> > - -- >> > Alexander Hansen >> > Fink User Liaison >> > -----BEGIN PGP SIGNATURE----- >> > Version: GnuPG v1.4.9 (Darwin) >> > Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ >> > >> > iEYEARECAAYFAkqK+dcACgkQB8UpO3rKjQ/5/ACfZlj4yC6vQ3koohwp+a8/sjdF >> > YBYAnRwM8jlk1vlNUu1ouDk51Di8LIsQ >> > =8Oat >> > -----END PGP SIGNATURE----- >> > > |