Compiling TEST_HEAD: IPO errors

Report or discuss software problems and other woes

Moderators: arango, robertson

Post Reply
Message
Author
nganju
Posts: 82
Joined: Mon Aug 16, 2004 8:47 pm
Location: U.S. Geological Survey, Woods Hole
Contact:

Compiling TEST_HEAD: IPO errors

#1 Unread post by nganju »

I installed MPICH2, and built the MCT libraries successfully (CYGWIN-ifort dual-core Windows computer). I used the same flags in my ROMS makefile as the MCT makefile. Compilation for TEST_HEAD goes smoothly until very near the end, where I get a plethora of "IPO errors". Previous postings on this subject seemed to deal with a the previous ROMS version (This is the roms 166 release).

A sampling is below:

Code: Select all

IPO Error: unresolved : _MPI_INIT@4
        Referenced in Build/master.o
IPO Error: unresolved : _MPI_COMM_SIZE@12
        Referenced in Build/master.o
        Referenced in Build/ocean_control.o
        Referenced in Build/ocean_coupler.o
        Referenced in libMODS.a(mod_parallel.o)
        Referenced in libSWAN.a(swanparll.o)
        Referenced in libSWAN.a(waves_coupler.o)
IPO Error: unresolved : _NETCDF_mp_NF90_STRERROR@12
        Referenced in libUTIL.a(close_io.o)
IPO Error: unresolved : _NETCDF_mp_NF90_SYNC@4
        Referenced in libUTIL.a(wrt_rst.o)
        Referenced in libUTIL.a(wrt_his.o)
        Referenced in libUTIL.a(wrt_info.o)
IPO Error: unresolved : _MPI_IRECV@32
        Referenced in libUTIL.a(mp_exchange.o)
        Referenced in libUTIL.a(distribute.o)
        Referenced in libSWAN.a(waves_coupler.o)
IPO Error: unresolved : _MPI_SEND@28
        Referenced in libUTIL.a(mp_exchange.o)
        Referenced in libSWAN.a(swanparll.o)
        Referenced in libSWAN.a(waves_coupler.o)
Any ideas are appreciated...

jcwarner
Posts: 1173
Joined: Wed Dec 31, 2003 6:16 pm
Location: USGS, USA

#2 Unread post by jcwarner »

unresolved : _MPI_INIT@4
unresolved : _MPI_COMM_SIZE@12
...
Check to see what MPI lib you are linking with. The MPICH2 distributed libraries have several flavors. The file README.winbin.rtf says:

• fmpich2.lib contains all caps cdecl: MPI_INIT
• fmpich2s.lib contains all caps stdcall: MPI_INIT@4
• fmpich2g.lib or fmpich2g.a contain lowercase cdecl: mpi_init__
which file are you linking with?

nganju
Posts: 82
Joined: Mon Aug 16, 2004 8:47 pm
Location: U.S. Geological Survey, Woods Hole
Contact:

#3 Unread post by nganju »

The MCT libraries were built with fmpich2.lib, and my makefile flag also uses fmpich2.lib. I have a feeling it may be a path issue: when I added the /usr/local/lib and /usr/local/include paths (where libmpeu.a and libmct.a are located), all unresolved externals went away except for the MPI-related ones.

nganju
Posts: 82
Joined: Mon Aug 16, 2004 8:47 pm
Location: U.S. Geological Survey, Woods Hole
Contact:

#4 Unread post by nganju »

After using the untouched-CSTM branch I created a few weeks back, and reinstalling MCT outside of the branch, I compiled with no problems. When I try to execute

Code: Select all

mpiexec -np 2 ./oceanM.exe ROMS/External/coupling_inlet_test.in
I get a uniquely USGS-type error...

Code: Select all

Credentials for nganju-pr rejected connecting to IGSWCAWWWS0835.gs.doi.net
Aborting: Unable to connect to IGSWCAWWWS0835.gs.doi.net
Is there an mpiexec option that would keep it from trying to connect this way? The "IGS..." identifier is my computer's name on the network...

nganju
Posts: 82
Joined: Mon Aug 16, 2004 8:47 pm
Location: U.S. Geological Survey, Woods Hole
Contact:

#5 Unread post by nganju »

found it... mpiexec -localonly....
it's running now...

Post Reply