Optimizations

While the algorithm presented in the previous section generates working code, we have found a number of optimizations that are useful to produce a more efficient simulation.

• Continuous assignment optimizations - Continuous assignments represent assignments to wires where the left hand side continuously reflects the current state of the variables on the right hand side. Rather than treating them as separate events we can inline them directly into the code, thus drastically reducing the number of events we need to schedule. Such inlining must be done in moderation, however, to limit the increase in code size. This optimization is performed on the event graph before scheduling takes place.

• Sensitization optimization - If an event immediately sensitizes itself after execution, and is sensitive at the beginning of simulation, it will always be sensitive. Thus it is not necessary to test for sensitivity in the generated code. In many models, this is the case for the majority of events.

• Levelization - During the course of event-driven simulation, it is possible to execute events multiple times as values propagate through the model. These multiple executions are not required for the correct answer and adversely impact simulation performance. Through proper ordering of event execution unnecessary events can be eliminated. This process is generally called levelization and is used in levelized compiled code simulators [2][1].

  One can add a form of levelization to an event-driven simulator by being intelligent about which events are retrieved from the event queue. We implement this in our compiler by assigning a level to each event in the graph based on its maximum length path from the start event. Then, when the function chooses events from , it chooses the event with the lowest level number.