Thanks for replaying Paul. I was counting valid lines every cycle so I was able to observe how the cache warming up process is taking place. It seems that when I use simmedium data set occupation of L2 caches is about 1MB (total) when approximately 50M instructions are comited. After this "warming up period", "occupation rate" simply stays on the same level until the end of simulation.
I run blackscholes with simlarge data set too, and similar situation happens but now caches contain 2MB max (it takes a little more time to achieve this fill in (maybe 80M commited instructions)). Is it possible that this is because the input size? but is it possible that this simmedium data set can warm up just 1MB cache. Similar situation happens with many other benchmarks with simmedium, but I would say that blackscholes is extreme case. Also, I noticed high hit rate for many benchmarks (PARSEC). Take blackscholes for instance and L1 hit rate (more than 99%). This is probably because all data are already in L1 and there is no need to demand other data? I don't know is everything what I am saying has any sense? I hope that I didn't say some crucial error:) Thanks again for replaying Paul. Like I said, any help is more than welcome. best regards Zoran
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
