A Comparative Study of the Precision of Stack Cache Occupancy Analyses

International audienceUtilizing a stack cache in a real-time system can aid predictability by avoiding interference between accesses to regular data and stack data. While loads and stores are guaranteed cache hits, explicit operations are required to manage the stack cache. The (timing) behavior of these operations depends on the cache occupancy, which has to be bounded during timing analysis. The precision of the computed occupancy bounds naturally impacts the precision of the timing analysis. In this work, we compare the precision of stack cache occupancy bounds computed by two different approaches: (1) classical inter-procedural data-flow analysis and (2) a specialized stack cache analysis (SCA). Our evaluation, using MiBench benchmarks, shows that the SCA technique usually provides more precise occupancy bounds