Abstract. This paper considers the automatic generation of architectural tests (ATGP), a fundamental problem in processor validation. ATGPs are complex conditional constraint satisfaction problems which typically feature both hard and soft constraints and very large domains (e.g., all memory addresses). Moreover, the goal is to generate a large number of diverse solutions under tight runtime constraints. To improve solution diversity, this paper proposes a novel approach to ATGPs by modeling them as MaxDiversekSet problems and solving them with constraintbased local search over conditional variables. The paper presents the semantics and implementation of conditional variables in this context and demonstrates the computational benefits of the approach.

No abstract

No abstract