Symbolic execution-based test generation is gaining popularity for software test generation. The increasing complexity of the software program is posing new challenges in software execution-based test generation because of the path explosion problem. We present a new 2-phase symbolic execution driven strategy that achieves high branch coverage in software quickly. Phase 1 follows a greedy approach that quickly covers as many branches as possible by exploring each branch through its corresponding shortest path prefix. Phase 2 covers the remaining branches that are left uncovered if the shortest path to the branch was infeasible. In Phase 1, a basic conflict driven learning is used to skip all the paths that may have any of the earlier encountered conflicting conditions, while in Phase 2, a more intelligent conflict driven learning is used to skip regions that do not have a feasible path to any unexplored branch. This results in considerable reduction in unnecessary SMT solver calls. Experimental results show that significant speedup can be achieved, effectively reducing the time to detect a bug and providing higher branch coverage for a fixed time out period than previous techniques. / Master of Science

Due to the widespread demands for efficient Propositional Satisfiability (SAT) solvers and its derivatives in Electronic Design Automation applications, methods to boost the performance of the SAT solver are highly desired. This dissertation aims to enhance the performance of SAT and related SAT solving problems. A hybrid solution to boost SAT solver performance is proposed as an initial attack in this dissertation, via an integration of local and DPLL-based search approaches. Next, a different hybrid strategy is attempted that takes advantage of the conflicts in the SAT search, which plays a critical role in modern SAT solvers. Usually a learned conflict-induced clause is added back to the clause database. Although conflict-induced clauses help to block a portion of the search space, they can also become a burden due to the added cost in memory consumption and Boolean Constraint Propagation (BCP). We thus propose a novel double-layer conflict-driven learning to store only those "primary" conflict clauses back into the clause database while keeping the other clauses as pseudo Boolean constraints. With this approach our experiments demonstrate that the approach can improve both in performance and memory consumption. This work opens the door on how to assess the usefulness of conflict induced clauses. Besides the aforementioned works about enhancing SAT solver performance and reducing memory cost, this dissertation also proposed a contributing work on the extended SAT problem solving. The current SAT solvers can provide an assignment for a satisfiable propositional formula. However, the capability for a SAT solver to return an "optimal" solution for a given objective function is severely lacking. MIN-ONE SAT is an optimization problem which requires the satisfying assignment with the minimal number of Ones, and it can be easily extended to minimize an arbitrary linear objective function. While some research has been conducted on MIN-ONE SAT, the existing algorithms do not scale very well on large formulas. This dissertation presents a novel approximation algorithm (RelaxSAT) for MIN-ONE SAT. RelaxSAT generates a set of constraints from the objective function to guide the search. The constraints are gradually relaxed to eliminate the conflicts with the original Boolean SAT formula until a solution is found. The experiments demonstrate that RelaxSAT is able to handle very large instances which cannot be solved by existing MIN-ONE algorithms; furthermore, RelaxSAT is able to obtain a very tight bound on the solution with one to two orders of magnitude speedup. Based on the proposed powerful MIN-ONE SAT algorithm, we built a MAX-SAT solver which achieved more than one order of magnitude speed up compared with the-state-of-art MAX-SAT solver. We also discuss a promising application of this MAX-SAT solver in formal verification. / Ph. D.

Backtracking CSP solvers provide a powerful framework for search and reasoning. The aim of constraint learning is increase global reasoning power by learning new constraints to boost reasoning and hopefully reduce search effort. In this thesis constraint learning is developed in several ways to make it faster and more powerful. First, lazy explanation generation is introduced, where explanations are generated as needed rather than continuously during propagation. This technique is shown to be effective is reducing the number of explanations generated substantially and consequently reducing the amount of time taken to complete a search, over a wide selection of benchmarks. Second, a series of experiments are undertaken investigating constraint forgetting, where constraints are discarded to avoid time and space costs associated with learning new constraints becoming too large. A major empirical investigation into the overheads introduced by unbounded constraint learning in CSP is conducted. This is the first such study in either CSP or SAT. Two significant results are obtained. The first is that typically a small percentage of learnt constraints do most propagation. While this is conventional wisdom, it has not previously been the subject of empirical study. The second is that even constraints that do no effective propagation can incur significant time overheads. Finally, the use of forgetting techniques from the literature is shown to significantly improve the performance of modern learning CSP solvers, contradicting some previous research. Finally, learning is generalised to use disjunctions of arbitrary constraints, where before only disjunctions of assignments and disassignments have been used in practice (g-nogood learning). The details of the implementation undertaken show that major gains in expressivity are available, and this is confirmed by a proof that it can save an exponential amount of search in practice compared with g-nogood learning. Experiments demonstrate the promise of the technique.