A symbolic execution framework for algorithm-level modelling and verification of computer microarchitecture

• Main Author: Hanna, Ziyad
• Format: Dissertation
• Language: English
• Published: University of Oxford 2011
• Subjects: Computer programs; Computer systems; Microprogramming; Verification

This dissertation addresses the challenge of modelling and functional verification for com- plex computer micro-architecture designs. It is evident that the emerging span and com- plexity of new computer architectures outstrip existing design modelling and verification methods. Several attempts in industry and academia, including High Level Modelling, still do not scale to address this problem. Typically they lack precise and clear language semantics for formal analysis, and do not have native support for concurrency, or the design language and methodology do not fit. Therefore, the gap between what current solutions provide and what industry needs is increasing. In this research we aim to leap ahead of the common incremental research in this area, and develop a new framework for algorithm level modelling and verification. We introduce a high level and executable Architectural Specification Language (ASL) for modelling the functional behaviour of the architectural algorithms. The semantics of our models is based on the theory of Abstract State Machines with synchronous parallel execution and finite choice, which we find naturally suitable for hardware modelling. Our framework is also powered by native symbolic execution algorithms for enabling high- level verification, design explorations and refinement checks of the high level models down to the design implementation. We developed a new framework that implements our ideas through ASL and supports symbolic execution. We demonstrate the utility of our language and symbolic execu- tion on examples and case studies in various modelling domains, and show a promising framework and methodology. We believe our approach will make it easier to explore micro-architectural algorithm behavior and easier to validate this using dynamic or formal techniques, thus yielding a promising attack on the modelling and verification problem, and enabling more productive convergence to high quality implementations.