> On Sept. 19, 2014, 3:40 p.m., Andreas Hansson wrote:
> > I'm just curious to know, do you think we will manage to eventually get the 
> > network power online again? It just seems unfortunate that we have to rely 
> > on post processing. I see how it is a step-wise modification, but perhaps 
> > it would be worth keeping parts of this if we will eventually go back to 
> > calling the power models while running.
> > 
> > We are about to add DRAMPower to ext to achieve exactly what I describe 
> > above. Ultimately I think it is better if the power numbers are available 
> > on-line. Less traces, less scripts, the possibility to do on-line 
> > power-aware decisions etc.
> 
> Brad Beckmann wrote:
>     Ideally that would be a great feature to have.  I think the key is to 
> avoid a lot of internal hacking of individual objects, like the code Nilay is 
> removing in this patch.  I'm not sure what the best final solution is.  Could 
> we dyncamically call the python scripts during simulation?  Could we somehow 
> directly incorporate the power model components into the sim object hierarchy?
> 
> Nilay Vaish wrote:
>     I think it is straight forward to invoke python scripts during 
> simulation.  We do it
>     all the time, don't we?  We can have pseudo instructions in the 
> application that
>     moves us back into the python hierarchy and then invokes the script.  Or 
> else, we can
>     have periodic intervals at which the simulator would return to python and 
> compute the
>     power consumption for the interval gone by.  I think both these can be 
> realized with 
>     some understanding how things work currently.  I think we should let the 
> users decide
>     how they want to do it.

I am not opposed this patch going in as is, but there are definitely reasons to 
avoid the C++ -> Python -> C++ ping-ponging.

#1 Performance. We have seen massive drops when using Python to do the power 
equation calculations. Of course it depends on granularity, but even in very 
realistic cases it is a massive problem.

#2 Dependency on Python. We are about to push a number of patches that allows 
gem5 to operate entirely without Python. This is important when using it hosted 
in other simulation frameworks e.g. SystemC or SST.


- Andreas


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://reviews.gem5.org/r/2404/#review5340
-----------------------------------------------------------


On Sept. 18, 2014, 9:23 p.m., Nilay Vaish wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://reviews.gem5.org/r/2404/
> -----------------------------------------------------------
> 
> (Updated Sept. 18, 2014, 9:23 p.m.)
> 
> 
> Review request for Default.
> 
> 
> Repository: gem5
> 
> 
> Description
> -------
> 
> Changeset 10373:f72ea8945c90
> ---------------------------
> ruby: network: garnet: remove functions for computing power
> 
> 
> Diffs
> -----
> 
>   src/mem/ruby/network/garnet/fixed-pipeline/GarnetNetwork_d.hh 4466307b8a2a 
>   src/mem/ruby/network/garnet/fixed-pipeline/GarnetNetwork_d.cc 4466307b8a2a 
>   src/mem/ruby/network/garnet/fixed-pipeline/NetworkLink_d.hh 4466307b8a2a 
>   src/mem/ruby/network/garnet/fixed-pipeline/Router_d.hh 4466307b8a2a 
>   src/mem/ruby/network/garnet/fixed-pipeline/Router_d.cc 4466307b8a2a 
> 
> Diff: http://reviews.gem5.org/r/2404/diff/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Nilay Vaish
> 
>

_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev

Reply via email to