Re: Value range tracing

jeremy@suede.sw.oz.au (Jeremy Fitzhardinge) writes:
|> I'm wondering if people bother with value range tracing. That is,
|> keeping track of the possible range of variables at each point in the
|> program. For example: (some stuff deleted)
|>
|> Is this kind of thing useful? Is it useful enough to implement?
|> This isn't very complex, so I assume people have thought about it,
|> but I've seen very little literature on it.


It is fully implemented in our automatic FORTRAN Parallelizer (PIPS).
See:


@INPROCEEDINGS{Irig:91,
AUTHOR = "Irigoin, Franc,ois
and Jouvelot, Pierre
and Triolet, Re'mi ",
TITLE = {Semantical Interprocedural Parallelization:
                                  An Overview of the {PIPS} project},
BOOKTITLE = International Conference on Supercomputing,
YEAR = 1991,
MONTH = jun,
}
(and our web pages: http://www.cri.ensmp.fr/pips).


It is an interprocedural extension of Cousot's algorithm:


@INPROCEEDINGS{Cous:78,
AUTHOR = {Cousot, Patrick and Halbwachs, Nicholas},
TITLE ={Automatic Discovery of Linear Restraints Among Variables of a Program},
BOOKTITLE = popl,
YEAR = 1978,
PAGES = {84--97}
}


In our parallelizer, it is used to enhance the dependence analysis, to
eliminate dead code, and to do some partial evaluation. But you may
think of many other applications. This type of information is
generally used to help subsequent analyses.


_ Beatrice CREUSILLET ___________ creusillet@cri.enmsp.fr ___


Centre de Recherche en Informatique, Ecole des Mines de Paris
35, rue Saint-Honore, F-77305 FONTAINEBLEAU Cedex FRANCE


Tel : +33 1 64 69 48 38 Fax : +33 1 64 69 47 09
                            http://www.cri.ensmp.fr/~creusil
--

Return to the comp.compilers page.
Search the comp.compilers archives again.