Traceability of Flow Information: Reconciling Compiler Optimizations and WCET Estimation

Real-time systems have become ubiquitous. For this class of systems, correctness implies not only producing the correct result, but also doing so within specified timing constraints. Designers are required to obtain the worst-case execution time (WCET) of their systems to guarantee that all applications meet their time constraints. Many WCET estimation methods have been proposed. They operate through static code analysis, measurements, or a combination of both. Such methods give an upper bound of the time required to execute a given task on a given hardware platform. To be useful, WCET estimates have to be as tight as possible. Information on possible flows of control (the so-called flow information) improves the tightness of WCET estimates. Flow information, should it be produced automatically or be inserted manually from annotations, is typically inserted at source code level. On the other hand, WCET analysis is performed at machine code level. Between these two levels, compiler optimizations may have a dramatic effect on the structure of the code, resulting in a loss of useful information. For this reason, many WCET tools for real-time systems turn off compiler optimizations when computing WCET. In this paper, we propose a framework to trace and maintain flow information from source code to machine code to benefit from optimizations, yet improving the WCET estimates. Our implementation in the LLVM compiler shows that we can improve the WCET of Mälardalen benchmarks by 60% in average, and up to 86%. We also provide new insight on the impact of existing optimizations on the WCET.