[mpich-discuss] Compilation problem (Dave Goodell)

[François PELLEGRINI]

Re Dave,

Dave Goodell <goodell at mcs.anl.gov> wrote:
> You can try the attached patch out against trunk at r4546 if
> you need to get unstuck right away.  I'll put something
> along these lines into the SVN trunk later today after
> a bit more testing and code review.

The patch worked perfectly, as far as compilation is
concerned. I'm gonna try out now how the version behaves
by running PT-Scotch.

BTW, there must be a dependency problem in the Makefiles.
When I compile the patched trunk with "make -j 3", I get :

make[1]: Entering directory `/home/local/packages/mpich2-r4546/examples'
  CC              cpi.c
../bin/mpicc   -o cpi cpi.o  -lm
cpi.o: In function `main':
cpi.c:(.text+0x54): undefined reference to `MPI_Init'
cpi.c:(.text+0x67): undefined reference to `MPI_Comm_size'
cpi.c:(.text+0x7a): undefined reference to `MPI_Comm_rank'
cpi.c:(.text+0x10b): undefined reference to `MPI_Bcast'
cpi.c:(.text+0x1a2): undefined reference to `MPI_Reduce'
/home/local/packages/mpich2-r4546/lib/libmpich.a(finalize.o): In function
`PMPI_Finalize':
/home/local/packages/mpich2-r4546/src/mpi/init/finalize.c:205: undefined
reference to `MPID_Finalize'
/home/local/packages/mpich2-r4546/lib/libmpich.a(finalize.o): In function
`MPIR_Add_finalize':
/home/local/packages/mpich2-r4546/src/mpi/init/finalize.c:56: undefined
reference to `MPID_Abort'

while it works fine with just "make".

With multi-core systems being the common rule, it is a bit
too bad that one cannot take advantage of them to speed-up
compilation.

Thanks again for the fix,


					f.p.