Thanks for catching this, Joel.

I'm curious how we didn't notice this... I'd think there'd be a significant
change in at least a few of our regression statistics.  Did it not make
much difference because our tests are lame, or did it make a difference but
we weren't paying attention?

Steve

On Tue, Aug 21, 2012 at 12:51 PM, Joel Hestness <[email protected]>wrote:

> Hey everyone,
>   I've found a pretty major bug in the Ruby memory controller that
> everyone should be aware of:
>
>   Back when the Ruby memory controller used the Ruby event queue interface
> (prior to changeset 8937:225590437eb2), the mem_bus_cycle_multiplier was
> used as the number of Ruby cycles before the memory controller should by
> cycled.  For example, if set to 10 (the default), the memory controller
> would cycle every 10 Ruby cycles, and by default the Ruby clock was set to
> 2GHz, so a multiplier of 10 would mean that the memory controller has an
> effective frequency of 200MHz.
>
>   In changeset 8937:225590437eb2, the memory controller was switched over
> to using the gem5 event queue interface, which uses ticks for scheduling
> events.  Despite the change from Ruby cycles to ticks for scheduling, the
> memory controller was still scheduling cycle events based on this
> multiplier value, simply adding it to the current tick.  Going back to our
> example of a multiplier of 10, this means that the memory controller would
> be cycled every 10 ticks or at an effective frequency of 100GHz if a tick
> is 1ps.
>
>   I wanted to give everyone a head's up about this bug, since I haven't
> seen anything on the email list previously.  I'm currently working on a
> fix, which I should have ready for review later today.
>
>   Joel
>
> --
>   Joel Hestness
>   PhD Student, Computer Architecture
>   Dept. of Computer Science, University of Wisconsin - Madison
>   http://www.cs.utexas.edu/~hestness
>
> _______________________________________________
> gem5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev

Reply via email to