[mpich-discuss] --enable-strictmpi not respected

Dettwyler, Levi E levi.e.dettwyler at intel.com
Wed Jul 20 16:18:01 CDT 2016

Hi there,

I think I've discovered an issue with the mpich-3.2 MPI tests not respecting the --enable-strictmpi flag passed in to the configure script. I'm building the mpich tests against OpenMPI.

The full configure line I'm using is given below:

export MPI_PREFIX=/home/build/OMPI
./configure --enable-strictmpi --disable-cxx --disable-fortran --prefix=$MPI_PREFIX CC=$MPI_PREFIX/bin/mpicc CFLAGS="-I$MPI_PREFIX/include" LDFLAGS="-L$MPI_PREFIX/lib64"

When running "make" in the mpich-3.2/test/mpi directory, the following error is generated:

Making all in ft
make[1]: Entering directory `/root/mpich-3.2/test/mpi/ft'
  CCLD     failure_ack
failure_ack.o: In function `main':
failure_ack.c:(.text+0x160): undefined reference to `MPIX_Comm_failure_ack'
failure_ack.c:(.text+0x1e3): undefined reference to `MPIX_Comm_failure_get_acked'
failure_ack.c:(.text+0x444): undefined reference to `MPIX_Comm_failure_get_acked'
collect2: error: ld returned 1 exit status
make[1]: *** [failure_ack] Error 1
make[1]: Leaving directory `/root/mpich-3.2/test/mpi/ft'
make: *** [all-recursive] Error 1

The compiler is picking up the declaration of MPIX_* from the mpich-3.2/src/include directory, but obviously the OpenMPI libs won't have definitions for those. However, the fact that it's searching for these symbols at all seems to indicate that the --enable-strictmpi option isn't being respected since the MPIX_* functions are nonstandard.

Any insight into this would be greatly appreciated!


Levi Dettwyler
Intel Corporation
levi.e.dettwyler at intel.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 3278 bytes
Desc: not available
URL: <http://lists.mpich.org/pipermail/discuss/attachments/20160720/ac46c1ac/attachment.p7s>
-------------- next part --------------
discuss mailing list     discuss at mpich.org
To manage subscription options or unsubscribe:

More information about the discuss mailing list