Hi,

I managed to create my own SimObject, and get that to compile/run into m5.
Now I want my SimObject track each pc on commit. How can I get m5 to call a
function in my object when the pc changes?  Is there some event based
mechanism for that, or should I just modify some of the source code in one
of the cores to directly call my object?

FYI, originally I was planning on making my module extend the trace system,
but eventually I want to build m5.fast, which I think disables tracing.
Also, its not really a trace mechanism (I'm not dumping out any stats), its
more of a control mechanism to decide when to offload exection to an
accelerator.


Andrew Lukefahr
[email protected]

Open Source, Open Minds


On Tue, Jan 25, 2011 at 4:00 PM, Gabe Black <[email protected]> wrote:

> You could do something like native trace but without actually modifying
> native trace. That system is modular, so you could make your widget
> completely separate and use it directly. Generally speaking, that seems
> like a pretty reasonable approach.
>
> Gabe
>
> Andrew Lukefahr wrote:
> > Hi,
> >
> > I guess I should have been more specific about the PC.  Static PC
> > dumps aren't quite what I want.
> >
> > The plan is to use a trace-cache type system to dynamically find hot
> > loops and offload them to an accelerator. For starters, I want to use
> > a sliding window to find small loops at runtime.  However, I'm not
> > exactly sure how to do that in m5.  Where would I put something like
> > that?  Can I hack some existing code, something like nativetrace, to
> > achieve that or would it be better to start something new? If I did
> > start something new, how do I integrate it into m5?
> >
> > Thanks
> >
> > Andrew Lukefahr
> > [email protected] <mailto:[email protected]>
> >
> > Open Source, Open Minds
> >
> >
> > On Wed, Jan 19, 2011 at 6:12 PM, Ali Saidi <[email protected]
> > <mailto:[email protected]>> wrote:
> >
> >     Hi Andrew,
> >
> >     ARM is still under active development, so you might run into
> >     issues, but there is a lot of stuff working.
> >
> >     SE mode works with the atomic simple, timing simple, and O3 cpu
> >     and runs SPEC benchmarks and other such things.
> >
> >     FS mode works enough to boot linux and run benchmarks with a
> >     atomic or timing simple cpu the o3 cpu doesn't work.
> >
> >     --trace-flags=ExecEnable should print the PC... if you would
> >     prefer a symbol you can add ExecSymbol.
> >
> >
> >
> >     Ali
> >
> >
> >
> >
> >
> >
> >
> >     On Wed, 19 Jan 2011 10:15:29 -0500, Andrew Lukefahr
> >     <[email protected] <mailto:[email protected]>> wrote:
> >
> >>     Hi,
> >>
> >>     I'm a grad student at the University of Michigan, and I'm looking
> >>     into using M5 for some research. I've got two questions:
> >>
> >>     1) Whats the current status of the ARM code? Are either the
> >>     syscall emulation or full system mode mode working currently?
> >>
> >>     2) For starters, I just want to dump the PCs to a file.  Is there
> >>     an easy way to find/print the current PC?
> >>
> >>     Thanks
> >>
> >>
> >>     Andrew Lukefahr
> >>     [email protected] <mailto:[email protected]>
> >>
> >>     Open Source, Open Minds
> >>
> >>
> >
> >
> >
> >     _______________________________________________
> >     m5-users mailing list
> >     [email protected] <mailto:[email protected]>
> >     http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
> >
> >
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > m5-users mailing list
> > [email protected]
> > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to