A Time Stamping Algorithm for Computing the Critical Path of Program Execution on Superscalar Processors

The increasing complexity of modern superscalar processors makes the evaluation of new designs more difficult. Current simulators such as Stanford’s SimOS [16] and the University of Wisconsin’s Simplescalar Toolset [2] perform detailed cycle-level simulation of the processor to obtain performance measurements at the cost of very slow simulation times. This report presents and analyzes an algorithm for computing the running time of a program running on an out-of-order superscalar by augmenting a fast in-order simulator with additional data structures for timestamping. The algorithm runs in constant time with respect to the issue width and fetch width, and in almost constant time (proportional to the inverse of Ackermann’s function) with respect to the size of the instruction window. A subset of these timestamping rules have been implemented in the SimOS simulator for evaluating the performance of several configurations of next generation superscalar processors [8].