Memory reuse analysis and transformation for ML-like programs

Abstract

Automatic memory management can be more than the garbage collection. Using the static analysis technologies, we can automatically find memory reuse opportunities in user programs, hence reducing the frequency of garbage collections. In this thesis, we present a static analysis that estimates reusable memory cells and a source-level transformation that inserts explicit memory-reuse commands into the programs that support algebraic data types. For benchmark programs in ML, our analysis and transformation achieves the memory reuse ratio from 5.2% to 91.3%. The small-ratio cases are for programs that have too much sharings among memory cells. For other cases, our experimental results are encouraging in terms of accuracy and cost. Major features of our analysis are: (1) use of multiset formulas in expressing the sharings and partitionings of heap cells; (2) poly-variant analysis of functions by parameterization for the argument heap cells; (3) deallocations conditioned by dynamic flags that are passed as extra arguments to functions. Our method is better than existing reported works in terms of its analysis accuracy and effectiveness. Our analysis and transformation is fully automatic and proven correct. Our experiment results are promising enough to conclude that garbage collection combined with the static analysis techniques can comprise more efficient, automatic memory management than garbage collection alone does.