[petsc-dev] Win32 and MPIAPI calling convention

[Satish Balay]

On Fri, 21 May 2010, Lisandro Dalcin wrote:

> I noticed that MPIAPI is being using for MPI callbacks, likely in
> order to suport Microsoft HPC Cluster Pack, right?

yes.

> Well, if you want to support DeinoMPI (yet another child of MPICH2),
> you have to use MPI_CALL.
> 
> Perhaps just:
> 
> #if defined(DEINO_MPI)
> #define MPIAPI MPI_CALL
> #endif
> 
> would do the trick.

If this indeed works - you can push it to petsc-31 repo..


> PS: Just FYI, MS MPI has a bug in its headers, if you ever need to
> call MPI_File_{c2f|f2c}, you can hostfix like this:
> 
> #undef  MPI_File_c2f
> #define MPI_File_c2f PMPI_File_c2f
> #undef  MPI_File_f2c
> #define MPI_File_f2c PMPI_File_f2c

Currently they are not used in petsc - so I guess we don't need them yet.

> However, I do not  know if MS MPI do actually have any support for any Fortran.

I does work with intel fortran compiler [with MS C]

Satish