[Octopus-devel] [Octopus-notify] svn commit: r3408 - in trunk: src/basic by xavier

[Xavier Andrade]

On Thu, 25 Oct 2007, Florian Lorenzen wrote:
>
> I would suggest that we take mpicc as the C compiler any time
> --enable-mpi is on the configure line, independent of
> --enable-libnbc. Then, we let configure check for the preprocessor
> which will amount to something like $(CC) -E, and $(CC) either being
> mpicc or cc or gcc, ... That means we have to move the MPI checks
> further up, I guess.
>

Right now we don't check at all for the mpi support in C, we should add it
(at least for the case that libnbc is enabled). This could also mean that we
could use parallel metis at some point.

But if the include of mpif.h is handled by the fortran compiler, there is no
need for mpi flags in the cpp and we can always use the CPP of autotools.

> The advantage of this is that we can use the CPP search of the
> autotools, otherwise, we have to do it ourselves.
>
> But we can still leave gthe inclusion of mpif.h to the Fortran
> compiler, as it actually is intended that way. (Is the include
> statement in the standard, btw.?)
>

I think include is standard Fortran 90, but if the compiler doesn't support
include, then the configure check for mpif.h will fail anyway.

> I can try so sort out the configure changes and check it on Soloris,
> which has the same problem as Mac OS X. Can you then check on Mac OS X?
>

Yes I can check it.

Cheers,

Xavier