On May 17, 2010, at 9:12 AM, Christopher Cambly wrote:

>
> Would it be possible to have this platform test Boost with MPI enabled,
> preferably with IBM's MPI?  Results for the parallel parts of Boost.Graph
> are useful to have on this compiler because of its use by many
> supercomputers.  Thank you.
>
 
The IBM MPI compiler is available on the machine we use for Boost result reporting. At one point we did experiment with enabling the Boost.MPI runs for IBM's MPI compiler however the Boost MPI toolset had trouble configuring the IBM MPI compiler. If I recall correctly the MPI toolset expected some type of compiler option that showed the location of the MPI headers and libraries but the IBM MPI compiler did not provide any options that proved useful.

In the short term, I was wondering if we should just modify our local copy of the mpCC script to print the information in a form that could be consumed by the Boost MPI toolset. Are the options some combination of: -showme:compile, -showme:link, -showme, -show, -compile_info and -link_info?

In parallel, I can work with the POE team to see if we can get something more official added to the IBM MPI wrapper scripts.

Hi Chris,

We added support to mpi.jam for IBM's POE.  It's a bit of a hack but the attached patch has worked for us, perhaps some variant of that would work for you?  Note that we explicitly rely on command matching failure in order to separately identify POE from, for example, Sun's (older) HPC library.

Please pass on to Mike that I'd still be quite pleased to run IBM Visual Age tests (if that would help).  I'd just need you to commit the changes to fix process_jam_log that we've talked about.

-- Noel