Re: [Gems-users] cpu-switch-time

[Dan Gibson <degibson@xxxxxxxx>]

With cpu-switch-time = 100, Simics will simulate each cpu for 100 cycles, then switch to the next processor. As to what memory model and how instructions appear to interleave, see the Simics documentation for details, or the Simics forums (www.simics.net).

For all of our simulations with GEMS, we set the cpu-switch-time to 1 to avoid these timing concerns. This occurs in the "gen-scripts" section, from mfacet.py if I remember correctly.

Regards,
Dan

arrvindh shriraman wrote:

Does cpu-switch-time influence timing aspects in ruby ?

In interface.C the rubyeventqueue is woken up for every cycle advanced on processor 0, to process events and advance global time. and in Sequencer.c->Message.C . The arrival time of a Load message is specified to be global event queue's current time.

Consider the following code with cpu-switch-time of 100.

Simics Quantum 1
Processor 1:
Ld
...
...

Processor 2:
Ld
...
==================
Does this mean processor2's load will only be seen at time - 100 since processor 1 has to complete . If there's another Ld from Processor 2 after this quantum has finished, what is the timestamp on that.

---

New Yahoo! Messenger with Voice. Call regular phones from your PC for low, low rates.

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users