RE: [mpi-21] ABI - call for working group

[Edric Ellis <Edric.Ellis@xxxxxxxxxxxxxxx>]

> >To respond briefly to some of the other points raised in this thread
> >from the perspective of The MathWorks - yes, something like a "morph"
> >layer would be acceptable - but to be effective for our customers,
this
> >needs to be something that they have ready access to (i.e. they don't
> >need to build it, as I believe they would with the current MorphMPI
> >solution). Compatibility of mpirun/mpiexec would certainly be useful
for
> >our customers, but that's easier to work around since no
recompilation
> >is required.
> Since you're distributing your software in binary form, you would
> compile the morph layer yourself and distribute your software already
> pre-linked with it. The morph layer would then adapt to your local
> MPI when you install your software. At install time it would need to
> have enough information about the local MPI to be able to link to it
> but shouldn't need much more than that.

But surely the morph layer needs to be compiled against the local MPI? I
would much prefer a situation whereby my customers do not have to build
anything. By having a standard binary interface (which MPI
implementations are free to satisfy using the morph layer - and there
should still be an option to compile and link against the "native"
implementation to avoid any performance penalty in the morph layer),
there's nothing to compile. 

Cheers,

Edric.