Thanks Jeff, these are very good questions!! I would start by us defining the "user" terminology, to put us on the same page. e.g., the user is it the developer using MPI (aka ISV) is it the end-user running the application... etc. Thanks, .Erez -----Original Message----- From: mpi3-abi-bounces_at_[hidden] [mailto:mpi3-abi-bounces_at_[hidden]] On Behalf Of Jeff Brown Sent: Thursday, February 21, 2008 8:37 AM To: mpi3-abi_at_[hidden] Subject: [Mpi3-abi] Fwd: questions to drive the telecon all, I put together some questions/comments that can help drive the telecon this morning. Jeff
ABI WG telecon agenga/notes/questions
- what is the motivation for this effort? ISVs cannot distribute a single binary that includes MPI code - must have either a translation layer or separate binaries for each mpi implementation (and compiler?) sites must maintain a separate "module" for every compiler/mpi combination on the system users must recompile for every compiler/mpi combination
- what are we trying to accomplish? app compiled with a "standard" mpi.h/mpif.h is compatible with standard compliant mpi implementations compiled on that same platform this would allow for run-time dynamic linking does not address items external to the binary such as environment variables and command line arguments
- what do we mean by an "ABI"
- comments on a translation layer may be somewhat viable for an ISV, but not acceptable for a user site just another layer of software to maintain may or may not track what we have on site
- what work is involved? "standard" mpi.h anything else?
- how do we integrate into the standard?
- what is the process to extend "standard" mpi.h content?
- what's the proposal process?
- what do we present at the March meeting?
_______________________________________________ Mpi3-abi mailing list Mpi3-abi_at_[hidden] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-abi