Autopack
Message packing for MPI made easy
Why
reinvent the wheel every time you write message-passing code?
Autopack
is a message-passing library which transparently packs small messages
into fewer larger ones for more efficient transport by MPI. It
can be used by any MPI program. The underlying message size is
easily adjusted at runtime for best efficiency. Other features of
Autopack include:
- collective commands that proceed asynchronously
- automatic management of MPI send and receive requests
- management of message buffer memory
- determination of the number of anticipated incoming
messages from a set of arbitrary sends
- deterministic message delivery from multiple sources to aid in
program testing or debugging
Autopack is written in C using MPI and should compile on any platform
with any implementation of MPI. Download is
available below.
We welcome questions, comments, requests for features, etc.
E-mail [email protected]
Coming Soon
Autopack v1.4 (already tested but
currently being documented) features a set of MPI-like drop-in
functions for ease of upgrading existing code. These provide
message packing through calls that mimic the immediate-mode sends and
receives (i.e. MPI_Isend(), MPI_Irecv()), as well as support for the
test/wait functions. In many codes, only a few lines
need to be changed to make use of Autopack.
Downloads
Mailing list
Announcements of new releases, bug
fixes, etc. will be posted to the
mailing list,
autopack-announce.
Subscribe by sending e-mail to
[email protected]
with the
message body "subscribe autopack-announce".
More info
Under
construction...
An overview presentation of Autopack v1.4 :
(pdf)
J. Flaherty, R. Loy, et al.
Load
Balancing and Communication Optimization for Parallel Adaptive Finite
Element Computation, Proc. XVII Int. Conf. Chilean Comp. Sci.
Soc., pp. 246--255, 1997.
Acknowledgments
This software project began under support from the Math and Computer Science Division of
Argonne
National Laboratory and the ASC / Alliances Center for
Astrophysical Thermonuclear Flashes. Current development is supported by the
Climate Change Research Program through the DOE Scientific Discovery through
Advanced (SciDAC) Computing program.
Last update: 16 June 2005