Re: [Gems-users] regd creating checkpoints for workloads

[Alaa Alameldeen <alaa@xxxxxxxxxxx>]

Chris,

When you build a benchmark checkpoint, you need to warm it up in two of phases. 
In the first phase, you need to get all the benchmark's files copied over, 
compiled, and run for any initialization period it may require to warm up 
memory. In the second phase, you need to run the benchmark till the caches are 
warmed up. For the first phase, you only need Simics running. You only need Ruby 
running in the second phase so that you can get a warmed-up cache.

The "load-caches" command in Ruby is used to load a cache warmup file after you 
have created the checkpoint and saved the cache using the "save-caches" command.

-Alaa

Christian Bienia wrote:
> Hi,
>
> I'm kind of confused: If you wanna create a checkpoint of a benchmark
> with warmed up caches, don't you have to load ruby/opal right from the
> beginning? Or does Simics manage the contents of the caches and
> ruby/opal just hook in? What else is ruby's load-caches parameter for if
> not for creating checkpoints?
>
> Chris
>
>
> On Tue, 2005-06-21 at 16:47, Mike Marty wrote:
>
> > Also you may want to look into using Simics in "fast mode", or
> > emulation-only mode for setting up checkpoints.  Supposedly the slowdown
> > is only 8-10x over a real machine.  However no timing modules can be
> > loaded in this mode.
> >
> > --Mike
> >
> >
> >
> > > Brinda,
> > >
> > > If you know how many instructions you want to run, you can continue simics for
> > > this same number of steps:
> > >
> > > c NUMBER_OF_INSTRUCTIONS
> > >
> > > and then take a checkpoint. You can load from this checkpoint and then run
> > > with Ruby/Opal for another number of steps. Please see the
> > > $GEMS_ROOT/gen-scripts/mfacet.py script for how we do it with magic
> > > instructions, but you can change the script to run for a certain number of
> > > steps.
> > >
> > > -Alaa
> > >
> > > Brinda Ganesh wrote:
> > >
> > > > Hi
> > > >
> > > > I want to create checkpoints and cache warmup dumps
> > > > for the spec cpu 2000 workloads past the
> > > > initialization points or after the initial 10 million
> > > > instructions.
> > > >
> > > > In order to create these checkpoints it seemed to me
> > > > that one would have to compile in magic instructions
> > > > for each benchmark which get triggered at start-up or
> > > > at an appropriate point. Then you step the correct
> > > > number of instructions and dump the cache.
> > > >
> > > > I wanted to know if there was a technique which would
> > > > not require using the magic instruction framework and
> > > > recompilation of the benchmarks.
> > > > For eg. is there is a way in the  simics interface/
> > > > opal interface - by which I can start benchmark run on
> > > > the simulated processor and run it for the appropriate
> > > > number of instructions. and thus not have to use the
> > > > os command line to start the application.
> > > >
> > > > Thanks in advance
> > > > Brinda
> > > >
> > > >
> > > >
> > > >
> > > > ___________________________________________________________
> > > > How much free photo storage do you get? Store your holiday
> > > > snaps for FREE with Yahoo! Photos http://uk.photos.yahoo.com
> > > > _______________________________________________
> > > > Gems-users mailing list
> > > > Gems-users@xxxxxxxxxxx
> > > > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> > >
> > > _______________________________________________
> > > Gems-users mailing list
> > > Gems-users@xxxxxxxxxxx
> > > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> >
> > _______________________________________________
> > Gems-users mailing list
> > Gems-users@xxxxxxxxxxx
> > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>
>
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users