On Tue, 14 May 2013, Andreas Hansson wrote:
On May 13, 2013, 3:55 p.m., Andreas Hansson wrote:
Overall it looks good to me. Are the resulting stats unchanged?
Nilay Vaish wrote:
I have not ran the regression tests yet. But obviously the statistics
files would change since new stats are being added to stats.txt and
several are being removed ruby.stats. I think the miss
counts should remain the same for the protocols where they were being
collected before. The hit counts were not being collected, so that would
be an addition.
Sounds good. It would be nice to know that values don't change (or that the
change is understood).
I ran the regression tests a while back. The results are same as before
except for the MOESI CMP directory protocol and for the MOESI hammer
protocol. Statistics for the MOESI CMP directory protocol were not being
recorded. So there is no easy way to check the stats being obtained now.
In case of MOESI hammer, the statistics for the L1 caches agree, while
those for the L2 cache don't. For one of the regression tests, I went
through the detailed statistics that are printed in the ruby.stats file
and confirmed that the statistics being generated now are correct and the
earlier ones were incorrect.
--
Nilay
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
