[mvapich-discuss] mvapich2-1.6rc1 problem (fwd)

[Bryan Putnam]

On Mon, 6 Dec 2010, Jonathan Perkins wrote:

> Can you give the output of `mpiname -a' with the mvapich2-1.5 version
> as well?  I'm wondering about the --with-pm selection.  It seems that
> you should not be setting this to remshell if you're trying to use
> mpirun_rsh.

Turns out that removing the "--with-pm..." did not fix the problem. If I 
build mvapich2-1.6rc1 with PGI compilers, everything appears OK. However 
if I build with Intel or gcc compilers, C programs will not run, although 
Fortran and C++ programs seem to be running OK.

Note that I don't see this problem with mvapich2-1.5, and have not seen 
the problem with any versions of mpich2 up to the current mpich2-1.3.1.

Thanks,
Bryan

> 
> On Mon, Dec 6, 2010 at 2:11 PM, Bryan Putnam <bfp at purdue.edu> wrote:
> > On Mon, 6 Dec 2010, Jonathan Perkins wrote:
> >
> >> I'm not sure what is causing your problem.  I've built your program
> >> with mvapich2-1.6rc1 on a couple of our systems and things are running
> >> fine.  Did you give any special configuration options during your
> >> build of mvapich2-1.6rc1?  Did you recompile your application with
> >> 1.6rc1?  What does `mpiname -a' output?
> >
> > Hi,
> >
> > Yes, I recompiled the code using the newly build mvapich2-1.6rc1.
> >
> > The results of "mpiname -a" give
> >
> > steele-adm 1069% mpiname -a
> > MVAPICH2 1.6rc1 2010-11-12 ch3:mrail
> >
> > Compilation
> > CC: icc -O3 -O3 -fpic -DNDEBUG
> > CXX: icpc -O3 -O3 -fpic -DNDEBUG
> > F77: ifort  -O3 -fpic -DNDEBUG -L/usr/lib64
> > F90: ifort  -O3 -fpic -DNDEBUG
> >
> > Configuration
> > --with-rdma=gen2 --with-ib-libpath=/usr/lib64 --enable-fast
> > --enable-threads --enable-debuginfo --enable-shared --enable-f77
> > --enable-f90 --enable-cxx --enable-romio --with-pm=remshell --without-mpe
> > --prefix=/apps/rhel5/mvapich2-1.6rc1/64/ib-intel-11.1.072
> >
> >
> > My configure options are:
> >
> > $MPI_SRC/configure \
> >  --with-rdma=gen2 \
> >  --with-ib-libpath=/usr/lib64 \
> >  --enable-fast \
> >  --enable-threads \
> >  --enable-debuginfo \
> >  --enable-shared \
> >  --enable-f77 \
> >  --enable-f90 \
> >  --enable-cxx \
> >  --enable-romio \
> >  --with-pm=remshell \
> >  --without-mpe \
> >  --prefix=$MPI_INSTALL/$CVER \
> >  > configure_$CVER.log 2>&1
> >
> >
> > Note that although this build was done using intel-11.1.072, I'm seeing
> > the same problem with gcc-4.4.0 builds. Note that my mvapich2-1.5 build is
> > still working correctly.
> >
> > Thanks,
> > Bryan
> >
> >
> >> > On Mon, Dec 6, 2010 at 10:52 AM, Bryan Putnam <bfp at purdue.edu> wrote:
> >> >
> >> >
> >> > I've found that one of the standard "ping pong" programs is failing to run
> >> > with mvapich2-1.6rc1. This code runs successfully with mvapich2-1.5. I'm
> >> > seeing the problem on both IB and iWARP systems.
> >> >
> >> >
> >> > steele-d000 1004% mpirun_rsh -hostfile $PBS_NODEFILE -np 2 ./pp2
> >> > MPI process (rank: 0) terminated unexpectedly on
> >> > steele-d000.rcac.purdue.edu
> >> > Exit code -5 signaled from steele-d000
> >> > MPI process (rank: 1) terminated unexpectedly on
> >> > steele-d001.rcac.purdue.edu
> >> >
> >> > The C code is attached.
> >> >
> >> > Thanks for taking a look,
> >> >
> >> > Bryan
> >> > _______________________________________________
> >> > mvapich-discuss mailing list
> >> > mvapich-discuss at cse.ohio-state.edu
> >> > http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
> >> >
> >> >
> >>
> >>
> >>
> >> --
> >> Jonathan Perkins
> >> http://www.cse.ohio-state.edu/~perkinjo
> >>
> >
> >
> >
> 
> 
> 
> -- 
> Jonathan Perkins
> http://www.cse.ohio-state.edu/~perkinjo
>