itaps-parallel Re: Question for Harold

[Tim Tautges]

[bounced because of non-member submission, so I'm re-sending]

Karin:

After looking at things and thinking about it I am fine with (mesh
instance, part handle) for referencing. Creating a whole new set of
interface functions seems like a lot of work for interface developers
and the applications. So if it can be avoided I certainly am for that.

Harold...

-----Original Message-----
From: Devine, Karen D [mailto:kddevin at sandia.gov]
Sent: Monday, February 11, 2008 10:08 AM
To: Trease, Harold E
Cc: itaps-parallel at mcs.anl.gov
Subject: Question for Harold


Hi, Harold.

The ITAPS Parallel Interface subcommittee has a question for you.

We have defined partitions to be assignments of entities to subsets, and
parts to be those subsets.  We assume that a mesh instance may have
several partitions, but a partition contains entities from only one mesh
instance.
We also assume a processor may have many parts, but each part is
contained completely within only one processor.

We identify partitions with a (mesh instance, partition handle) pair.
That is, functions that operate on partitions would take a mesh instance
and a partition handle as input arguments.

Similarly, within a processor, we would like to identify parts with a
(mesh instance, part handle) pair so that we can overload many of the
iMesh EntitySet functions with part handles.  That is, we would like to
allow, say, iMesh_getNumOfType to accept either an entity set handle OR
a part handle, and return the number of entities of a given type in the
entity set or the part, respectively.  There was some concern, however,
that (1) some implementations may not be able to distinguish between
part and entity set handles, and (2) some implementations might not be
able to uniquely identify parts given only the mesh instance and part
handle (e.g., they might also need the partition handle).

The alternative to overloading the EntitySet argument is to define many
new interface functions that take the mesh instance, partition handle,
and part handle as input and duplicate the EntitySet functionality on
parts.

Within the subcommittee, we agreed that we could accept the (mesh
instance, part handle) pair only if your implementations could support
it.  So, you wield the power now.  :)  Would the (mesh instance, part
handle) pair be sufficient for uniquely identify parts on a processor?

Thanks.
Karen





-- 
================================================================
"You will keep in perfect peace him whose mind is
   steadfast, because he trusts in you."               Isaiah 26:3

              Tim Tautges            Argonne National Laboratory
          (tautges at mcs.anl.gov)      (telecommuting from UW-Madison)
          phone: (608) 263-8485      1500 Engineering Dr.
            fax: (608) 263-4499      Madison, WI 53706