Compiler Support for Measurement-based Timing Analysis

Abstract

Measurement-based timing analysis (MBTA) techniques have been developed as a complimentary to static WCET analysis, in order to exploit worst-case execution time (WCET) analysis at early stages of system development. The direct advantage of MBTA is that, in contrast to static WCET analysis, no timing model of the hardware platform has to be developed. Instead the timing model is generated automatically by performing systematic execution time measurements. MBTA provides high retargetability, as the test suite used for execution time measurements is typically derived from the source code of the program. In order to provide an accurate WCET estimate, the test suite has to provide a sufficient coverage of the temporal system behavior. Here also the compilation tool chain is important as the compiler may introduce additional control flow that is not visible at the source code. In this paper we present FORTAS, an MBTA tool that systematically generates test data using a range of different techniques, like heuristics and model checking. Furthermore, we show how compiler support for MBTA can provide code optimization while preserving the code coverage achieved by the MBTA test suite at source-code level. First evaluations indicate that the performance penalty for ensuring coverage preservation of the test suite is low