Worst Case Execution Time Estimation for Advanced Processor Architectures

TL;DR: A measurement based approach to estimate the worst case execution time on a fully featured processor, suitable for processors equipped with advanced acceleration techniques, and an existing extreme value statistic approach is extended, to handle combinations of measurements.

Abstract: Advanced acceleration features, as they are used in todays mass market, high performance processors, have only been considered in isolation in previous worst case execution time estimation approaches. This thesis presents a measurement based approach to estimate the worst case execution time on a fully featured processor. To produce reliable results several aspects have to be considered. Prior to the start of a measurement, the acceleration techniques are preset, as far as possible, into their worst case state. The features, which cannot be controlled to produce the worst case state are either randomised or covered by penalties added to the measured results. All possible path combinations are enforced using additional instrumentation code. By partitioning the measurement problem into several measurement blocks, the coverage of all path combinations is ensured. To cover final uncertainty, an existing extreme value statistic approach is extended, to handle combinations of measurements. Additionally a scheduling analysis method, suitable for processors equipped with such acceleration techniques, is presented. A number of test cases, show the applicability and the limitations of the approach.