On Mon, Jan 2, 2012 at 12:05 PM, Nilay Vaish <[email protected]> wrote:

> As of now, I am not too keen on this inter-cpu checkpoint restore. Let's
> first start with restoring a checkpoint taken using TimingSimpleCPU or O3
> to itself.
>

I think you're actually making the problem harder, not easier.  Our
intention has always been that a checkpoint should only contain
architectural state, so that you can generate a checkpoint from one model
and restore it into a different one.  So getting a CPU model other than
AtomicSimpleCPU to both generate and restore from a checkpoint is two
problems, while getting it to simply restore from a checkpoint is a strict
subset of that.


> So should we have a command line argument that specifies what cpu to
> assume for restoring from the checkpoint?
>

In the long run, we should just assume that the CPU model you've asked for
is the one that is restored from the checkpoint.  The
restore-to-atomic-then-switchover-even-when-you-didn't-ask-for-it behavior
is a hack and should go away when it's no longer needed.  It should still
be possible to restore to an atomic CPU, run for a specified period of time
to warm up the caches, then switch to a more detailed model, but that's
something you should specifically have to request on the command line.

Maybe in the interim there does need to be another command line argument to
enable or disable the current hack, but that should just be a temporary
thing.

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

Reply via email to