The working set sizes of blackscholes and bodytrack are relatively small, so it is likely to see most of the cache lines remain invalid (never touched), especially when you have a large cache capacity.
You can find working set sizes of each Parsec benchmark (with input=simlarge) in this paper: http://parsec.cs.princeton.edu/doc/parsec-report.pdf As of how to properly warm up CPU caches, what we do is simply run more instructions. Just to remind, the number of instructions executed usually refers to the total instructions committed by all of the cores, so if you let Marss simulates a 4 core processor for 1 billion instructions, then each core only commits 250 million instructions. Depending on the number of cores and the cache capacity per core, you might need to adjust the number of instructions to simulate. On Sat, Nov 23, 2013 at 12:00 PM, <[email protected]>wrote: > Send Marss86-Devel mailing list submissions to > [email protected] > > To subscribe or unsubscribe via the World Wide Web, visit > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > or, via email, send a message with subject or body 'help' to > [email protected] > > You can reach the person managing the list at > [email protected] > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Marss86-Devel digest..." > > > Today's Topics: > > 1. Re: cache time to warm up (Paul Rosenfeld) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Sat, 23 Nov 2013 10:05:45 -0500 > From: Paul Rosenfeld <[email protected]> > To: Zoran Jaksic <[email protected]> > Cc: marss86-devel <[email protected]> > Subject: Re: [marss86-devel] cache time to warm up > Message-ID: > < > cag_sxbpr8pxbkfgqrnfeyj85o+dqfqa6kqepa-zux9zql2o...@mail.gmail.com> > Content-Type: text/plain; charset="iso-8859-1" > > Interesting. When you dump your log file, is that an instantaneous snapshot > at cycle 1.5B or are you writing the log every cycle? > > > On Fri, Nov 22, 2013 at 10:36 AM, Zoran Jaksic <[email protected]> wrote: > > > Hi guys, > > > > I have one small question. If anyone could help me I would be very > > grateful. Please answer if you have any hint. > > > > I run blackscholes benchmark from the image you provided. I used script > to > > generate checkpoint and I used simmedium input set. I used moesi default > > configuration for the machine. Only thing that I changed in the code is > > some small part that counts the number of valid lines of every cache and > > prints them in log file. however when I run this simulation until the end > > it seems that more than 80% of the cache lines are invalid all the time > (in > > L3) and every L2. L2 are 2MB and L3 8MB but more than 1.5 billion > > instructions are commited, so it should be enough? Similar situations is > > with many other parsec benchmarks. Do you have any clue what is happening > > here. Is this normal, or is there any way to warm up cache before > > simulation? > > > > Thank you very much in forward > > Best regards > > Zoran > > > > > > > > _______________________________________________ > > http://www.marss86.org > > Marss86-Devel mailing list > > [email protected] > > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > > > > > -------------- next part -------------- > An HTML attachment was scrubbed... > URL: < > https://www.cs.binghamton.edu/mailman/private/marss86-devel/attachments/20131123/48d6d7fd/attachment-0001.html > > > > ------------------------------ > > _______________________________________________ > http://www.marss86.org > Marss86-Devel mailing list > [email protected] > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > > > End of Marss86-Devel Digest, Vol 45, Issue 9 > ******************************************** >
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
