[PRL] Wasowski @ Harvard -- Fri 9/3
Mitchell Wand
wand at ccs.neu.edu
Tue Aug 31 10:19:47 EDT 2004
From: Greg Morrisett <greg at eecs.harvard.edu>
Sender: church-announce-bounces at types.bu.edu
To: church-announce at types.bu.edu
Subject: [Church-announce] Andrzej Wasowski speaks on Code Gen. for Embedded
Systems
Date: Thu, 26 Aug 2004 12:45:52 -0400
Andrzej Wasowski, a PhD student at the IT Univ. of Copenhagen
working with Peter Sestoft will be visiting Harvard on Friday,
Sept. 3rd. He'll be giving a talk in Maxwell Dworkin 221 at
11am on his work on code generation for resource-constrained
embedded systems. He's been working on how to translate
statechart models into efficient C programs, and writes:
One of the more interesting theoretical results of my work is a
process algebraic framework for specifying behavioral specialization
scenarios for synchronous reactive systems.
[snip]
Our framework (developed together with prof. Kim G. Larsen in his
model-checking group at Aalborg University in Denmark) goes further
than elimination of dead code, introducing the notion of
color-blindness to environment models. Our environments are not only
restricted in the number of traces they can produce, but also in the
number of system answers they can distinguish. It may well be that an
environment can execute all parts of the embedded system, but simply
ignores the granularity of the answers (for example it may wait for a
result, not caring about the actual value). This color-blind
disability changes dynamically depending on the environment state. We
have found that color-blindness can model surprisingly many realistic
situations, for example causality, boolean switches, disconnected
actuators or unified actuators. On the theoretical level
color-blindness is achieved by relaxing synchronization mechanism of
processes from label equality to dynamically changing coarser
equivalence relations. It shall allow new kinds of optimizations,
based not only on availability of inputs, but also on relevance of
outputs.
-Greg
More information about the PRL
mailing list