HOME       UP       PREV       FURTHER NOTES       NEXT (TLM - Measuring Utilisation and Modelling Contention)  

TLM Modelling: Adding Timing Annotations Continued

The preferred coding style is more flexible: we pass a time accumulator variable called `delay' around for various models to augment where time would pass (clearly this causes far fewer entries to the SystemC kernel):

  // Preferred coding style
  putbyte(char d, sc_time &delay)  // The delay variable records how far ahead of kernel time this thread has advanced.
    delay += sc_time(140, SC_NS); // It should be increment at each point where time would pass...

But, at any point, any thread can resynch itself with the kernel by performing

   // Resynch idiomatic form:
   delay = 0;

Important note: Simulation performance is reduced when there are frequent resynchs, but true transaction ordering will be modelled correctly.

At a parallel fork, we need to make a fresh accumulator variable and at any join we continue with the larger delay (ignoring 'eureka' joins).

23: (C) 2008-11, DJ Greaves, University of Cambridge, Computer Laboratory.