[Acpc-l] Vortrag Dr. Jens KNOOP, 23.5.2000
Maria Cherry
Maria Cherry <maria@par.univie.ac.at>
Wed, 17 May 2000 12:46:02 +0200 (MET DST)
UNIVERSITAET WIEN
INSTITUT FUER SOFTWAREWISSENSCHAFT
FWF-Projekt Spezialforschungsbereich F011 "AURORA"
EINLADUNG ZU EINEM VORTRAG IM RAHMEN DES AURORA-KOLLOQUIUMS
Parallel Data-Flow Analysis of Parallel Programs
Dr. Jens KNOOP
University of Dortmund
Germany
ZEIT: Dienstag, 23. 5. 2000, 10.00 Uhr c.t.
ORT: Institut fuer Softwarewissenschaft
1090 Wien, Liechtensteinstrasse 22,
Seminarraum, Mezzanin
Abstract
A major obstacle of analyzing parallel programs results from the
combinatorial explosion of the number of interleavings manifesting
their semantics. Current research indicates that there is no general
way for by-passing the `state explosion problem', however, we could
show that specific problem classes including the practically most
important class of GEN/KILL-analyses, are an exception. They can be
performed as easily and efficiently as their sequential counterparts.
In this talk, we will show that these analyses have naturally
parallelizable counterparts. Central for this is to complement the
approach they are derived from with an approach for demand-driven DFA,
and -- under the perspective of program verification -- the derivation of
strongest postconditions conventional DFA is aiming at from weakest
preconditions. Together this yields the key for constructing interactive
tools like debuggers as well as parallel versions of conventional and `hot
spot' optimizers for the large and in practice widely used class of
GEN/KILL-based optimizations. We demonstrate this in detail for the
class of GEN/KILL-analyses, and discuss extensions as well as
limitations of the overall approach.