If you're talking about the simulator aspect, MARSS uses its own decoder
in order to produce additional information necessary for full-system
simulation.

When MARSS is not active, however, we still use qemu's translation
mechanism as we only need to translate the guest instruction streams to
native code.

Tyler

> Or maybe I got that backwards -- does marss read the pre-translated
> instruction stream?
>
> On Mon, Mar 25, 2013 at 4:34 PM, Paul Rosenfeld
> <[email protected]>wrote:
>
>> Thanks for the reply.
>>
>> So I guess it lets the TCG do a better job of creating optimized code
>> sequences for your CPU? But that *does* impact the simulation side of
>> things since the instruction stream that results is then run through
>> marssx86, correct?
>>
>> I'd be willing to help out with doing some cleanup on the scripts
>> (create_checkpoints.py could definitely use some basic features like
>> optparse and run_bench.py could use some sanity checks on making sure
>> the
>> checkpoint exists and how many cores/megs of memory it was created
>> with).
>>
>>
>>
>> On Mon, Mar 25, 2013 at 3:23 PM, Brendan Fitzgerald <
>> [email protected]> wrote:
>>
>>> Paul,
>>>
>>> After searching the -cpu option performs better with an architecture
>>> most
>>> similar to the one you're machine has. It helps to provide features
>>> specific to your cpu.
>>>
>>> Look in qemu/target-i386/cpuid.c for the list of cpu's provided and the
>>> available options.
>>>
>>> Brendan
>>>
>>> On Mon, Mar 25, 2013 at 2:34 PM, Paul Rosenfeld
>>> <[email protected]>wrote:
>>>
>>>> I noticed that by default, the create_checkpoints.py file passes the
>>>> flag '-cpu core2duo' to QEMU. Does this have any impact on the
>>>> checkpoint
>>>> creation process? Should I be passing the same flag when I run the
>>>> checkpoints in marss?
>>>>
>>>> I ask because I've noticed sometimes I get checkpoints that behave
>>>> strangely (i.e., random intervals of inactivity in the benchmark that
>>>> will
>>>> go away randomly if I re-checkpoint). I'm wondering if there's
>>>> something
>>>> about the checkpoint creation that might be causing this
>>>> non-determinism.
>>>>
>>>> Thanks,
>>>> Paul
>>>>
>>>> _______________________________________________
>>>> 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

Reply via email to