On Aug 19, 2004, at 3:21 AM, jnt7 wrote:
> However, only mpif77 appears in my lam bin directory, and obviously
> the F90
> bindings were not made.
> If I use FC=xlf90, then F77 bindings won't be made... (correct?).
>
> MPICH allows one to make bindings for both; does anyone know if this is
> possible in LAM as well?
LAM/MPI does not have a complete Fortran 90 interface. Among the
pieces missing are the mpif90 wrapper compiler, the Fortran module
interface, and some of the specialized kind functions.
Keep in mind that MPI doesn't really have "F77 bindings" and "F90
bindings". The Fortran bindings are officially written to Fortran 90,
but in a way kind to Fortran 77 compilers. The big difference between
F77 and F90 from the MPI point of view is the availability of the MPI
module for Fortran 90.
If your application does not use the F90 MPI module, then you can most
likely use LAM's Fortran interface without problem for both Fortran 77
and Fortran 90 (especially if you use the XL compilers for both f77 and
f90). If your application does require the F90 MPI module, then LAM
does not have the Fortran support you require at this time. We are
working on it, and intend to have full Fortran 90 compliance in the
near future.
As for how to build the thing, there are really two options - you can
try building LAM with FC=xlf90 and try using the F90 compiler
everywhere (specifying mpif77 for both the F77 and F90 compiler). Or
you can use FC=xlf and use mpif77 -showme to determine which flags to
specify for the Fortran 90 compiler and avoid the wrapper compiler for
F90 code. We generally prefer users avoid doing this, as sometimes we
have to change the flags in the wrapper compilers.
I hope this helps...
Brian
--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/
|
