Furthering Datalog in the pursuit of program analysis

This thesis focuses on specifying compiler analyses in a declarative language, namely Datalog. This work contains three contributions: expanding the number of programs Datalog, specifically Soufflé (a superset of Datalog), can support; providing a declarative implementation of a well-known analysis; and introducing a new program analysis. First, a new construct is introduced into Soufflé allowing consistent negation of relations used to infer facts about that same negated relation. This can be seen as a join point in inference when a rule partially pauses inference until the negated relation is in a consistent state, and it is safe to evaluate the negation. Soufflé, with consistent negation, is then used to specify a natural encoding of a depth-first graph exploration using a visited set and the sum of affine multivariable polynomials with partially disjoint variable domains. The technique allows previously difficult-to-implement problems to be implemented with asymptotically better performance, in a more natural and expressive fashion. Second, a declarative implementation of GVN is presented—a common program analysis to find equivalent program variables—called DGVN. This implementation breaks a data dependence throughout the whole GVN algorithm that previously limited its parallelism. The Soufflé implementation also makes use of the previously described new negation-aiding construct to increase performance and to allow a natural expression of the algorithm. DGVN is complete enough to be evaluated on benchmarks from Specint. Third, a definition of pairwise program commutativity is presented, useful to automatic parallelising compilers that transform data dependency graphs. This definition is then lifted into the set domain, allowing for interface-boundary respecting commutativity specifications using semantic program properties, as source-code inline annotations. This system is then used to encode commutative functions from libc and other common data structures and employed to parallelise several benchmark applications.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2022

Author dc:creator

• Isaacs, Joseph

Advisors dc:contributor.advisor

• Jones, Timothy
• Mycroft, Alan

• computer science

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/0552a9e4-156d-4e96-a8ab-73ab9e2ab582/download
• https://creativecommons.org/licenses/by-sa/4.0/

Language dc:language

eng

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.100114

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/354249