Hi Abhishek,

Getting superscalar inorder to work will most likely require some
minor tweaks. I say "most likely" because I tried it with
multiprogrammed workloads comprised of spec benchmarks on a
multithreaded in-order CPU and it did not work. There were minor bugs.
I got it working with very specific tweaks particular to what I wanted
to model. My changes are specific to dispatching multiple instructions
(at most one instruction from a one hardware thread) in an SMT
configuration. Your are using splash benchmarks and seem to be using
an in-order core without any multithreading support.

Korey, I think I should be able to find out one of my patches in which
the stagewidth can be set from python (I replaced some static arrays
with dynamic arrays). However, it doesn't make much sense to just add
that patch unless I fix the bugs that surface thereafter. I will try
to take out some time to work on it.

-Soumyaroop

On Fri, Apr 23, 2010 at 1:51 PM, Korey Sewell <[email protected]> wrote:
> Hi,
> the inorder model is still under development and the superscalar
> functionality hasnt been tested yet. However, any feedback and debugging
> help you can provide would be much appreciated (please send coding questions
> to m5-dev...)
>
> In regards to your current question about stagewidth, that is in artifact of
> hard-coded values being used and the python and C++ not talking to each
> other correctly at configuration time.
>
> Try changing the stageWidth parameter noted in InOrderCPU.py and see what
> you get. Note that,  the hard-coded values in pipeline_traits.hh,cc are used
> to defined the interstage buffers and the like so you would also need to
> adjust that to match your python config OR setup M5 so that the python param
> "stagewidth" is writing to the namespace visible stagewidth defined in
> pipeline_traits.hh.
>
> If you do the latter, please provide the patch to m5-dev :)
>
> Also, since it hasnt been tested I'm not terribly confident it will work out
> the box, but any issues you run into , please send to m5-dev mailing list
> and we can try to help you sort them out if you are willing to put in a
> little coding work as well.
>
> Thanks!
>
> -Korey
>
> On Fri, Apr 23, 2010 at 1:14 PM, abhishek rawat <[email protected]> wrote:
>>
>> Hi,
>> I have been trying to change the stageWidth parameter in inorder CPU model
>> (ALPHA SE mode) and I keep getting the following from the simulation output:
>>
>> Exiting @ cycle 9223372036854775807 because simulate() limit reached
>>
>> I am running splash2's Radix benchmark.
>>
>> Also, even after setting "stageWidth" to a value different than 1 in
>> pipeline_traits.hh and re-compiling, the stageWidth in the generated
>> stats.txt remains 1. Is it possible that I could be doing something wrong.
>>
>>
>> Thanks,
>> Abhishek
>>
>> _______________________________________________
>> m5-users mailing list
>> [email protected]
>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
>
>
> --
> - Korey
>
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>



-- 
Soumyaroop Roy
Ph.D. Candidate
Department of Computer Science and Engineering
University of South Florida, Tampa
http://www.csee.usf.edu/~sroy
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to