PTL logo

From: Ben Martin (benjmart_at_[hidden])
Date: 2008-10-22 13:13:29


Hmmm. Two things occur to me. The first is, be sure to rerun configure
when changing which MPI implementation you're building against. I assume
you're probably already doing that, but, just in case...

The second is that while in theory MPI.NET should just build against any
MPI, in practice there's no way to be sure in advance. I know that it
works with OpenMPI and MPICH2. If when you say MPICH you mean MPICH2,
then it should have built. You might try it with OpenMPI just to see if
it builds.

If it's failing against OpenMPI or even MPICH2, then I'll probably need
more information to figure out the problem, since I'm able to build both
on Linux without a problem. Could you tell me more about the system,
like distro and architecture? You might also check the version of perl
since it's a perl script that generates CustomUnsafe.cs.

Ben Martin
Open Systems Laboratory

Jo Shields wrote:
> Hi,
>
> I'm trying to compile MPI.NET on Linux, but hitting problems.
>
> It seems the generated CustomUnsafe.cs file isn't happy about compiling,
> against any of my test MPI implementations (SGI MPT, InfiniPath, MPICH),
> throwing 39 different CS0283 errors:
>
> CustomUnsafe.cs(106,31): error CS0283: The type `System.IntPtr' cannot
> be declared const
>
> These are caused by definitions in that file of constants with type
> MPI_Comm, MPI_Datatype, MPI_Group, MP
> I_Info, MPI_Errhandler, MPI_Op, or
> MPI_Request. None of these types are real, of course - they're defined
> as shorthand for System.IntPtr earlier in the file. Constants can't be
> pointers.
>
> Any suggestions?
> _______________________________________________
> mpi.net mailing list
> mpi.net_at_[hidden]
> http://www.osl.iu.edu/mailman/listinfo.cgi/mpi.net
>