[Firm] libfirm Erweiterung um IR bäume zu durchsuchen

Andreas Zwinkau zwinkau at kit.edu
Fri Nov 15 09:05:11 CET 2013


Am 14.11.2013 17:04, schrieb Matthias Braun:
> The closest thing in the literature is probably the graph transformation
> community. We used to have a lot of research in that direction in the
> form of the grgen tool (I guess Sebastian can tell you more about that).
> The last thing I heard though is that the graph transformation tools
> still don't beat the manually programmed rules in libfirm, because the
> manually programmed rules make use of the fact, that we have a huge
> number of patterns we search concurrently and all patterns start a
> distinguished root node, these assumptions are not made by general graph
> transformation systems from what I understood.

Graph transformation tools always match general graphs. Most patterns
are trees though, which can be match more efficiently. One fundamental
question is, if a pattern matcher should support trees, DAGs, or graphs
in general.

Furthermore, as far as I know, there is no graph transformation tool,
which matches "any from lots of small patterns". Usually, you have to
search for one pattern after another. I had an idea (based on RETE
algorithm) how to tackle this problem, if someone wants to approach it.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 555 bytes
Desc: OpenPGP digital signature
URL: <http://lists.ira.uni-karlsruhe.de/pipermail/firm/attachments/20131115/99a61e2d/attachment.sig>


More information about the Firm mailing list