Model-driven development of pipelined systems

点击翻译按钮获取中文摘要

Pipelining is a widely used performance optimization for digital-hardware systems that is used in systems ranging from simple signal-processing filters to high-performance microprocessors. Pipelining increases performance by overlapping the execution of multiple instructions, analogous to the way that multiple cars flow through an automobile assembly line. Although intuitive when viewed as block diagrams, pipelines are difficult to design and verify, because subtle bugs can arise from the interactions between instructions flowing through the pipeline. The goal of this research is to reduce the development effort and improve the quality of digital-hardware systems by creating theories, tools, and techniques for the model-driven development of pipelined systems. Model-driven development is a style of development that focusses on models of a system, rather than just the final implementation. Model-driven development distinguishes between a model of an individual system and a metamodel that models an entire class of systems. A metamodel captures features that are common to all systems in the class and identifies points of variation between systems. We make daily use of metamodels without being burdened by the meta-level nature of our thinking, or even being aware that we are using a model. Conventions for cartographic symbols on road maps are an example: all areas are drawn at the same scale, water is blue, thicker lines mean larger roads, etc. These conventions are a metamodel for maps, and an individual map is a model for a specific geographical area. We will create a family of metamodels that describe the behaviour of pipelined systems in general and of specific performance optimizations for pipelines. We will encode these metamodels in CAD tools that will enable engineers to reuse high-level design strategies and cleanly separate the process of designing a system from that of implementing the design. The research will improve design quality by reducing the opportunities to make mistakes and making it easier to discover opportunities for new performance optimizations.