Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] F90 open-mpi module bug
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2011-05-21 14:23:14


On May 21, 2011, at 10:20 AM, Dan Reynolds wrote:

> Rightly so. Sorry for posting a bug to the developer list. It's just that the bug was years old and had compiled on so many systems, plus the gfortran error message was entirely unhelpful, so I hadn't realized to look at the arguments as the cause of the error. My guess is that the F90 module in other distrubutions (MPICH, MVAPICH, IBM, ...) don't declare "intent" on the variables, so it always just went through.

Or they don't have explicit interfaces for all MPI functions. If they don't have an explicit interface for MPI_ABORT, then it follows the same implicit rules that otherwise apply in Fortran (e.g., what all mpif.h interfaces I know do).

Indeed, OMPI defaults to only having explicit interfaces for MPI functions that do not take "choice" buffers (e.g., MPI_INIT, but not MPI_SEND) because of various complicates due to that (that will be changing sometime in the not-distant future, though -- stay tuned!).

-- 
Jeff Squyres
jsquyres_at_[hidden]
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/