On Fri, May 18, 2012 at 10:01 AM, Xin Tong <[email protected]>wrote:
> so it is using the rdtsc() to determine the time ?
>
> In simulation mode, when 'rdtsc()' is executed (in VM) Marss returns:
offset + simulation_cycles executed
Where 'offset' is the clock at the start of simulation and
'simulation_cycles' executed are around 200K cycles per second of real-time.
Thus the clock in VM is really slow because it depends on how much
'simulation' cycles are completed.
- Avadh
Thanks
>
> Xin
>
>
> On Fri, May 18, 2012 at 12:50 PM, Paul Rosenfeld <[email protected]>wrote:
>
>> Yes, simulated wall time is much much slower than real wall time. If
>> you're running at at say 2GHz, the simulator has to simulate 2 billion
>> cycles for each of the n cores you're running per second. 4 minutes is 480B
>> cycles -- which is a lot of cycles to simulate.
>>
>> I hope I'm not misunderstanding your question.
>>
>> On Fri, May 18, 2012 at 12:41 PM, Xin Tong <[email protected]>wrote:
>>
>>> it seems that the timing on the simulated processor is slower than wall
>>> clock time as well. i.e. one of the bencmarks i am running runs itself for
>>> 4 mins. but the on the simulated ooo core processor, 4 mins seems to be
>>> much longer then 4 mins wall clock time.
>>>
>>>
>>> Thanks
>>>
>>> Xin
>>>
>>>
>>> _______________________________________________
>>> http://www.marss86.org
>>> Marss86-Devel mailing list
>>> [email protected]
>>> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>>>
>>>
>>
>
> _______________________________________________
> http://www.marss86.org
> Marss86-Devel mailing list
> [email protected]
> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>
>
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
