Hello
If you are using the PARSEC suite from the University of Texas
(http://www.cs.utexas.edu/~parsec_m5/), then you should read their report
(http://www.cs.utexas.edu/~parsec_m5/TR-09-32.pdf) to understand how they
customized the benchmarks. They basically divided each application in 3
sections: an initial serial phase, the parallel phase (or ROI - region of
interest) and a final serial phase. They used a special library to dump stats
at the end of each phase. The m5 dumpstats and m5 resetstats commans in your
rcS file are there to ensure you reset (and dump) all stats before the
benchmark's execution.
In this case, the use of the --fast-forward is a small trick to speedup the
booting of the system. The number of instructions to be fast forwarded
(1000000000) is higher than the actual number of instruction executed while
booting, that is why the actual switching is triggered from the rcS file (m5
switchcpu command) right before the start of the benchmark. The assumption you
make at the end of your e-mail is true, however with --fast-forward 1000000000
is am almost sure that you pass the boot and reach the benchmark.
Best regards,
Alexandru Iordan
________________________________
From: Heba Saadeldeen <[email protected]>
To: gem5 users mailing list <[email protected]>
Sent: Wednesday, March 7, 2012 9:41 PM
Subject: [gem5-users] Statistics file
Hi,
I have two questions regarding running Parsec on Gem5
I am using the following command to run it without checkpoints:
./build/ALPHA/gem5.opt --stats-file=bodytrack2-try.txt ./configs/example/fs.py
--script=/cs/arch2/heba/haha/gem5/Parsec/bodytrack_2c_simsmall.rcS --num-cpus 2
--cpu-type detailed --caches --l2cache --fast-forward 1000000000 --maxinsts
100000000
And my script is:
cd /parsec/install/bin
#/sbin/m5 switchcpu
#/sbin/m5 dumpstats
/sbin/m5 resetstats
./bodytrack /parsec/install/inputs/bodytrack/sequenceB_1 4 1 1000 5 0 2
echo "Done :D"
/sbin/m5 exit
/sbin/m5 exit
1- Why the statistics output has 3 begin/end simulation section? which one to
look at? I deliberately commented dumpstats such that it does not write the
statistics for the part before the actual execution.
2- What if I am using switchcpu in the script. Does that mean it will do
functional, and then transition to detailed simulation at beginning of program
execution ignoring the fastforward. Or does it start the fastforward after the
switching?
I assumed that by my script (not using switchcpu) the simulation will run in
functional simulation until the fastforward ends and then switch to detailed
whether or not you reached the actual program execution. Is that true?
Any help really appreciated!
Thanks,
--
Heba
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
