Interpreted graph models

Abstract

A model class called an Interpreted Graph Model (IGM) is defined. This class includes a large number of graph-based models that are used in asynchronous circuit design and other applications of concurrecy. The defining characteristic of this model class is an underlying static graph-like structure where behavioural semantics are attached using additional entities, such as tokens or node/arc states. The similarities in notation and expressive power allow a number of operations on these formalisms, such as visualisation, interactive simulation, serialisation, schematic entry and model conversion to be generalised. A software framework called Workcraft was developed to take advantage of these properties of IGMs. Workcraft provides an environment for rapid prototyping of graph-like models and related tools. It provides a large set of standardised functions that considerably facilitate the task of providing tool support for any IGM. The concept of Interpreted Graph Models is the result of research on methods of application of lower level models, such as Petri nets, as a back-end for simulation and verification of higher level models that are more easily manipulated. The goal is to achieve a high degree of automation of this process. In particular, a method for verification of speed-independence of asynchronous circuits is presented. Using this method, the circuit is specified as a gate netlist and its environment is specified as a Signal Transition Graph. The circuit is then automatically translated into a behaviourally equivalent Petri net model. This model is then composed with the specification of the environment. A number of important properties can be established on this compound model, such as the absence of deadlocks and hazards. If a trace is found that violates the required property, it is automatically interpreted in terms of switching of the gates in the original gate-level circuit specification and may be presented visually to the circuit designer. A similar technique is also used for the verification of a model called Static Data Flow Structure (SDFS). This high level model describes the behaviour of an asynchronous data path. SDFS is particularly interesting because it models complex behaviours such as preemption, early evaluation and speculation. Preemption is a technique which allows to destroy data objects in a computation pipeline if the result of computation is no longer needed, reducing the power consumption. Early evaluation allows a circuit to compute the output using a subset of its inputs and preempting the inputs which are not needed. In speculation, all conflicting branches of computation run concurrently without waiting for the selecting condition; once the selecting condition is computed the unneeded branches are preempted. The automated Petri net based verification technique is especially useful in this case because of the complex nature of these features. As a result of this work, a number of cases are presented where the concept of IGMs and the Workcraft tool were instrumental. These include the design of two different types of arbiter circuits, the design and debugging of the SDFS model, synthesis of asynchronous circuits from the Conditional Partial Order Graph model and the modification of the workflow of Balsa asynchronous circuit synthesis system.