../

Meeting With Kevin

  • Optimizing texada
    • Initially took a week and crashed and 100G
    • Find out the signal sources and eliminate aliases
    • Original traces has 100,000
    • removing stuck at 0 reduces to 38000 and grouping reduces to 7000 groups
      • Based on just the trace
      • The rules are based on the group rather than the signal
      • 100M rules for that trace
      • Groups from RTL vs Empirical groups
      • missing RTL groups means coverage
      • This could be another feedback loop HOME journal for this kind of publications

Mine

  • Though it may be interesting to create an open-source version of commercial tools, it may not be appropriate for us
  • Run SVA vs LTL benchmark

Grouping

  • Yasas’s grouping works on the traces, and only on the traces. It is a preprocessing step to make Texada run faster
  • RTL analysis is another plane. It will result in a different set of groups that the traces. The traces merely compare values at each timestep so we may have aliases that are never intended/impossible to detect with SA
  • So naturally there is a gap between the runtime behavior and the design intent. This could be another feedback loop for Periscope

Trace comparison

  • How to most get the distinct signals from a set of signals?
    • Hash them?
    • Brute force seems to be $n^2$ in the number of signals