Hi all,

 

I have implemented a stream prefetcher in m5, and run the simulation in SE
mode. However, the prefetcher seems to perform much better than previous
works' statistics (these work usually uses trace simulators). For example,
cfp2000.art gives almost no miss in the whole simulation. And I am wondering
why this is happening .

 

Any suggestions?

Is SE mode using virtual addresses? And because these virtual addresses are
continuous without interrupts such as different pages in FS mode, the stream
prefetcher is more likely to get good performance?

 

Thanks a lot in advance!

 

Yours,

Lunkai

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

Reply via email to