[MPI3-IO] Topics for discussion for the I/O working group slot

Rob Ross rross at mcs.anl.gov
Tue Feb 26 08:51:13 CST 2013


You might consider a non-collective version (name TBD) that does not allow for changing the data rep (i.e., doesn't have that parameter) and that makes it erroneous to use shared file pointer routines after the call, until after a following collective file set view (this because you can't know if all file views are sync'd or not, and probably they are not).

Many hints would be ignored by this non-collective version, but I think that's fine.


On Feb 26, 2013, at 8:21 AM, Rob Latham wrote:

> On Mon, Feb 25, 2013 at 05:26:28PM -0600, Mohamad Chaarawi wrote:
>> Hi Dries,
>> On 2/25/2013 3:44 PM, Dries Kimpe wrote:
>>> - Fix shared file pointer access cfr. the EuroMPI paper?
>> yes thanks for reminding me :-)
>>> - independent file view changes
>> Do you think, implementation wise, that it would be straightforward
>> to just remove this restriction? I'm worried how that would effect
>> shared file pointer operations in the standard.
>> I'm all for having a new routine to change the file view
>> independently with restrictions on what comes after that.
> Yes, please.  MPI_File_set_view_not_dumb would be a fine name.
> Or one could introduce a crapton of routines that take a file view in
> addition to a memory datatype and obviate the need for setting file
> view at all (well except for data representation but who actually does
> that?)
> ==rob
> -- 
> Rob Latham
> Mathematics and Computer Science Division
> Argonne National Lab, IL USA
> _______________________________________________
> MPI3-IO mailing list
> MPI3-IO at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io

More information about the mpiwg-io mailing list