Can you send the output of laminfo, the config.log file, and the
output from configure?
Also, can you run with the rpi_verbose variable set to 1000? E.g.:
mpirun -ssi rpi_verbose level:1000 ....
That might shed a little more light on what is happening. For
example, one cause for them failing is trying to run too many
processes on a single node, thereby running out of shared memory (the
User's Guide gives details on how much shared memory is used).
On Aug 16, 2005, at 2:36 PM, Grismer Matthew J Civ AFRL/VAAC wrote:
> We’re building lam-7.2b1r10193 under OS X 10.4.2 on a PowerMac G5,
> and using the XL Fortran compiler for Fortran support. We got the
> XL Fortran compiler working with mixed C/Fortran code (CC was the
> default gcc, which is version 4) by adding –lSystemStubs and /usr/
> lib/libgcc_s.1.0.dylib to the gcc_libs lines of /etc/opt/ibmcmp/xlf/
> 8.1/xlf.cfg. Lam was configured with the following:
>
>
> FC=xlf
>
> FFLAGS=-O -qarch=auto -qtune=auto -qcache=auto
>
> CFLAGS=-fast
>
>
> ./configure --with-rpi=sysv --with-memory-manager=none --disable-tv-
> queue
>
>
> The last configure option to disable TotalView support was required
> to get the build to complete, so that was the first issue. The
> second issue is that neither the sysv or usysv shared-memory rpi’s
> work. When you attempt to run any code with either of these lam
> prints the following message:
>
>
> ----------------------------------------------------------------------
> -------
>
> The selected RPI failed to initialize during MPI_INIT. This is a
>
> fatal error; I must abort.
>
>
> This occurred on host shelob.flight.wpafb.af.mil (n0).
>
> The PID of failed process was 24081 (MPI_COMM_WORLD rank: 0)
>
>
>
> Choosing another rpi, such as lamd or tcp, appears to work fine.
> Any suggestions?
>
>
> Matt
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
