On Feb 10, 2014, at 8:51 AM, David Radice <[email protected]> wrote:

> Hi Ian,
> 
> On Feb 10, 2014, at 3:42 AM, Ian Hinder <[email protected]> wrote:
> 
>> Has anybody run into this before?  If not, then I can attempt to dig into 
>> the configure script and find out why it's not detecting openmpi from 
>> macports anymore.
> 
> yes I ran into this problem.
> 
> The reason why the MPI thorn does not find the OpenMPI from MacPorts is that 
> MacPorts recently changed the way in which MPI is installed. They wanted to 
> support the simultaneous installation of multiple OpenMPI versions compiled 
> with different compilers, so they decided to move all the header files and 
> libraries of OpenMPI to some nonstandard locations where the MPI thorn does 
> not find them. If I am not mistaken the new paths look like these:
> 
> the headers in ${PREFIX}/include/${OMPI_VERSION} ,
> the libraries in ${PREFIX}/lib/${OMP_VERSION} .
> 
> The default MPI wrappers have also been renamed as ${OMPI_VERSION}-${NAME}. 
> The new port "mpi select" can create aliases for those, but since ET does not 
> use the MPI wrappers to compile, this does not help very much.
> 
> Possible solutions would be to either compile with the MPI wrappers, change 
> to option list to explicitly set MPI_INC_DIR and MPI_LIB_DIRS, or manually 
> compile OpenMPI and point the MPI thorn to it. I would recommend the last of 
> these options, because I am not sure that OpenMPI is compiled correctly by 
> MacPorts: it used to be the case that OpenMPI was compiled with a mix of 
> clang (for the C part) and gcc (for the Fortran code), which resulted in some 
> very weird behavior when running with OpenMPI and OpenMP, and I am not sure 
> that the new "gcc versions" of OpenMPI are actually compiled with gcc or if 
> they simply use gcc as the default target for the wrapper scripts.
> 
> Cheers,
> David

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

_______________________________________________
Users mailing list
[email protected]
http://lists.einsteintoolkit.org/mailman/listinfo/users

Reply via email to