Varia: Optimization by Logic Programming

DSpace/Manakin Repository

Varia: Optimization by Logic Programming

Citable link to this page

 

 
Title: Varia: Optimization by Logic Programming
Author: Redwine, Kevin; Heffner, Kelly

Note: Order does not necessarily reflect citation order of authors.

Citation: Redwine, Kevin and Kelly Heffner. 2005. Varia: Optimization by Logic Programming. Harvard Computer Science Group Technical Report TR-13-05.
Full Text & Related Files:
Abstract: We have designed a prototype compiler optimization infrastructure called Varia and demonstrated its potential to explore the space of optimizations. In Varia everything is represented as logic: instructions and blocks are axioms, analyses and transformations are inference rules, and optimization proceeds by forward-chaining deduction. Varia makes adding and combining optimizations simple—every rule may be eligible to fire at each step, so optimizations are automatically combined. We have coded five optimizations in our system and demonstrate that Varia can combine them. We describe how to translate normal iterative data-flow analyses into Varia, briefly discuss the performance of the rules engine, and finally, propose changes in the engine to make it more suitable for optimization. Things we need to edit: Massage an example so we can fully demonstrate how hypothetical facts allow the combination of two analyses/optimizations. Ie. How the hypothetical fact allows us to trigger an optimization, which then discharges the hypothesis, thus reaching a fixed point. I would really be happy if we could find some strong relation to logical axioms for this. I need to read all my modal logic books.
Terms of Use: This article is made available under the terms and conditions applicable to Other Posted Material, as set forth at http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAA
Citable link to this page: http://nrs.harvard.edu/urn-3:HUL.InstRepos:25620082
Downloads of this work:

Show full Dublin Core record

This item appears in the following Collection(s)

 
 

Search DASH


Advanced Search
 
 

Submitters