Section: New Results

WCET estimation for parallel code

Participant : Dumitru Potop Butucaru.

This is joint work with Isabelle Puaut, Inria, EPI ALF.

Classical timing analysis techniques for parallel code isolate micro-architecture analysis from the analysis of synchronizations between cores by performing them in two separate analysis phases (WCET – worst-case execution time – and WCRT – worst-case response time analyses). This isolation has its advantages, such as a reduction of the complexity of each analysis phase, and a separation of concerns that facilitates the development of analysis tools. But isolation also has a major drawback: a loss in precision which can be significant. To consider only one aspect, to be safe the WCET analysis of each synchronization-free sequential code region has to consider an undetermined micro-architecture state. This may result in overestimated WCETs, and consequently on pessimistic execution time bounds for the whole parallel application.

The contribution of this work [56] , [44] is an integrated WCET analysis approach that considers at the same time micro-architectural information and the synchronizations between cores. This is achieved by extending a state-of-the-art WCET estimation technique and tool to manage synchronizations and communications between the sequential threads running on the different cores. The benefits of the proposed method are twofold. On the one hand, the micro-architectural state is not lost between synchronization-free code regions running on the same core, which results in tighter execution time estimates. On the other hand, only one tool is required for the temporal validation of the parallel application, which reduces the complexity of the timing validation toolchain.

Such a holistic approach is made possible by the use of deterministic and composable software and hardware architectures (many-cores with no cache sharing and time-predictable interconnect, static assignment of the code and data to the memory banks). Such code can be written by hand or automatically synthesized using the Lopht tool 5.4 or other automatic parallelization techniques.