Hello All,

I have noticed that the CPI I get when I simulate a benchmark with X86 O3,
Full System, and Ruby memory system is very high compared to the CPI in
published research for the same benchmarks under similar configuration.

Here is an example of what I am talking about.

Benchmark      gem5 CPI     CPI in published research
mcf                   30.64              11.8
hmmer              1.22                ~8

I am calculating CPI from the stats file using numCycles / commitedInsts.

I wanted to verify that numCycles is getting updated correctly. Can someone
give any suggestions on how I can go about this?

As of now I am thinking of walking through the Event Queue insertions for a
few events (in sim/eventq.cc) and see if I can find something out of the
ordinary. I mean like a cache request / miss event scheduled later than it
is supposed to, which might be increasing the numCycles more than it is
supposed to, thus resulting in the high CPI. Is there any place else I can
check this?

Thanks in advance.

Akhil
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to