An API for dynamic partial evaluation under DynamoRIO

Abstract

Dynamic optimization systems have available runtime (dynamic) data and can cross more boundaries than traditional static optimization systems in the pursuit of program transformations. However, dynamic optimization systems are limited by the fact that any time spent in analysis or transformation is included in the running time of the program; thus any transformations must not only improve the performance of the program, but also make up for the time spent in analysis. Interpreters in general confound current heuristics for dynamic optimization. In previous work, the dynamic optimization system DynamoRIO was extended with Dynamo-RIO-with-Log-PC to address this issue by maintaining extra state; this thesis generalizes the additions made by DynamoRIO-with-Log-PC and develops a useful API for a wider range of programs running under the DynamoRIO system.