On Wed, 10 Sep 2014, Zi Yan via gem5-dev wrote:
OK. I got this quiesceCycles part.
My questions are
1) if CPU is in quiesce state, it should stop fetching instructions, right?
why total instructions differ so much? before is 2.1 billion, after is 3.5
billion.
I am speculating that you end the simulation once you reach a particular
point in the application you are running. I think the application is
performing a similar number of instructions in both cases. It is the
operating system is that is doing more work in the 'after' case than it
does in the 'before' case.
2) why changing Movfp and Pop87 op_class from FloatAddOp to IntAluOp
makes such big gap? What is the connection between those two instructions
and system.cpu.quiesceCycles?
As I understand, that changeset is only making a small timing difference.
But this small timing difference brings into motion other effects that
ultimately lead to big difference in the execution. Take a look at the
fourth set of statistical output. There also the two executions diverge
from each other in terms of time. But then they converge again. If you
run longer, may be for another 10 seconds on the target, then the two runs
might converge again.
--
Nilay
Thanks.
--
Best Regards
Yan Zi
On 10 Sep 2014, at 10:14, Nilay Vaish via gem5-dev wrote:
I have only looking at the first set of numbers all the time. Sorry about
that. So here is possible explanation. You run long enough that the
operating system would kick in multiple times. It seems the OS is doing
something in the middle of the simulation like waiting for an interrupt to
happen. Look at the value of system.cpu.quiesceCycles. There is a big
gap between the values for the two runs. I think the processor is not
doing much when it is waiting for an interrupt.
--
Nilay
On Wed, 10 Sep 2014, Zi Yan via gem5-dev wrote:
> Hi Nilay,
>
> Did you download the after file at: https://db.tt/5QgJ5BPH
> and before file at: https://db.tt/iqWdeu2K ?
>
> I guess it takes sometime for Dropbox to upload the whole file.
> I dump stats every 100 million instructions, so after file is 6.4MB
> where before file is 3.8MB.
>
> Please download them again. Sorry for the inconvenience.
>
> Thanks.
>
> --
> Best Regards
> Yan Zi
>
> On 10 Sep 2014, at 0:26, Nilay Vaish via gem5-dev wrote:
>
> > The value of sim_insts is 100,000,000 (after) and 100,000,001
> > (before). I
> > think you have uploaded files different from what you wanted to
> > upload.
> > Check the files that you have in your dropbox.
> >
> > --
> > Nilay
> >
> >
> > On Tue, 9 Sep 2014, Zi Yan via gem5-dev wrote:
> >
> > > Hi Nilay,
> > > > 1) Please do "vimdiff before_op_class_stats.txt >
> > after_op_class_stats.txt"
> > > 2) Then search for sim_insts.
> > > 3) Use shift+n, you will find out that before one only has 2.1
> > billion
> > > instruction simulated, whereas after one has 3.6 billion
> > instructions.
> > > > --
> > > Best Regards
> > > Yan Zi
> > >
> > _______________________________________________
> > gem5-dev mailing list
> > [email protected]
> > http://m5sim.org/mailman/listinfo/gem5-dev
> _______________________________________________
> gem5-dev mailing list
> [email protected]
> http://m5sim.org/mailman/listinfo/gem5-dev
>
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
