Re: Warren Abstract Machine

Andrew Gavin <acg@aethos.co.uk> wrote:
>[Info on the Warren Abstract Machine? Efficent unification & backtracking.]


Here's some papers I found interesting; they only describe WAM to
introduce VAM, but since you mention efficiency, they may be
appropriate.


One contribution of these papers is in showing how to statically bound
the effects of new code (rules) that are introduced dynamically.
There's other good stuff here, too.


I haven't looked for these lately, I don't know if the URLs are still
valid.


;-D on ( Lost in the Warren ) Pardo


%A Andreas Krall
%A Thomas Berger
%T A Progress Report on Incremental Global Compilation
of Prolog
%J Proceedings of the ILPS'94 Workshop on Implementation Techniques
for Logic Programming Languages
%D 1994
%W http://www.complang.tuwien.ac.at/andi/articles.html
%X (abstract) ``Traditional native code generating Prolog compilers
with global analysis compile programs as a whole and do not support
the data base builtin-predicates \fCassert\fP and \fCretract\fP. In
this paper we ... enable global analysis and incremental compoilation.
    ``... and expanded on the fly to machine code'' (see [KB 95]).


%A Andreas Krall
%A Thomas Berger
%T The VAM(AI) \- an Abstract Machine for Incremental
Dataflow Analysis of Prolog
%J Proceedings of the ICLP'95 Workshop on Abstract Interpretation
%D 1995
%W http://www.complang.tuwien.ac.at/andi/articles.html
%X The virtual machine code is used to drive abstract interpretation.
The vitual machine code is updated during abstract interpretation to
hold results of analysis.


%A Andreas Krall
%A Thomas Berger
%T Incremental Global Compilation of Prolog with the
Vienna Abstract Machine
%J Proceedings of the ICLP'95 Workshop on Abstract Interpretation
%D 1995
%W http://www.complang.tuwien.ac.at/andi/articles.html
%X Prolog allows program modification via \fCassert\fP (insert program
code) and \fCretract\fP (delete program code)
In general, adding code invalidates compiler assumptions, e.g. if some
variable types can become more general (e.g., integer and atom instead
of atom). Incremental analysis allows for recompilation of just the
parts where it is necessary.
--

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