Hi Gabe,
thank you very much for your answer. After that i was able to solve a
lot of problems.
The Parsec-Benchmark seems to run now fine. I get the statistics in
the m5out folder, and also a log of the system. Everything fine so far.
My Problem now is, that i just get statistics for the L1 and L2 cache
etc, but not for the memory, and unfortunately i need exactly this
statistics. How can i start the benchmark with ruby support, or so, so
that i get this statistics?
Another problem is the runtime for me. I used the checkpointing system
and also the m5.opt, but it takes very long time to pass the
benchmarks. Because m5 just use 1 core, i looked after a possebility
to use more cores. The wiki and the mailinglist is not 100% clear. I
have found a comment from 2008 or so, that there are plans to make
gem5 multithreaded. But also found some comments, that gem5 just can
use 1 thread.
Have somebody experiences whot is the best system to run gem5 very
fast? Whot makes gem5 fast? Large L3 cache? High IPC, lots of RAM, a
RAM-Disk, a SSD or something else?
Finally, are there problems, when i start more than one session of the
benchmark? i think, it would be helpful to start 10 sessions of the
parsec-benchmark with different cache and memory stats. So i can see
how they effect the results.
Btw. whot happens, when the benchmark run out of RAM? Can gem5 handle
this in FS mode with a virtual HDD?
Stephan Walter
-------- Original-Nachricht --------
Datum: Sat, 17 Sep 2011 19:50:52 -0400
Von: Gabriel Michael Black <[email protected]>
An: [email protected]
Betreff: Re: [gem5-users] Problem with gem5 FS-mode and parsec
A lot of the regression tests require you to have binaries that we
aren't allowed to distribute because of how they're licensed (part of
benchmark suites, for instance). Those tests will fail unless you have
our particular binaries because the tests are very picky, expecting
exactly a certain number of instructions to be executed with exactly
the right statistics as far as branch prediction, cache behavior, etc.
It's *extremely* unlikely that you would be able to generate those
binaries yourself in such a way that the tests will fail. If that's
what the problem is, then everything is actually fine. If the hello
world test works then you're probably ok. This is a recognized and
regularly commented on deficiency in our testing framework, and one of
these days we're going to fix it.
If you use fs.py, you don't have to (or at least shouldn't have to)
rename your binaries. You can use arguments to the script to change
what files it looks for.
If you want to see output from the console, you should use the m5term
program to connect to the system's simulated serial port. In FS mode,
the system doesn't print anything to the console you run it from. All
the output is directed over the simulated serial device like it would
be (or at least could be) in a real system. There should be
instructions on how to do that on the wiki.
I'm not that familiar with how the Parsec stuff is set up to work, but
if it's actually running the benchmarks it could take a really long
time. Simulated time passes much slower than real time when the
simulated system is actually doing something. I'd recommend you use
gem5.opt. It behaves basically the same, it just has optimization
turned on in gcc which can make a big difference. If something is
breaking and you're trying to debug the gem5 binary with gdb,
valgrind, etc., then gem5.debug might be the better choice since
functions won't be inlined, variables won't be optimized away, etc.,
but for running normally gem5.opt is a lot better.
Gabe
Quoting Stephan Walter <[email protected]>:
> Hi,
>
> i try to use gem5 for a small project at the university. Since 4
> weeks i try to run parsec on gem5, but i am not abel to do this.
>
> Ok first what i have done.
>
> A: Download gem5.tarball -> ok no good idea...
> B: get the gem5.stable repository from the server with mercurial
> C: get the FS-files from the gem5 download page
> D: get the Parsec-2.1 disk img, .rcS Script and the Linux 2.6.27
> kernel img from http://www.cs.utexas.edu/~parsec_m5/
> E: follow the instruction in the manual for the parsec benchmarks.
> Now there is my first problem. The kernel.org page is offline. I
> think this is not a problem, because of the linux 2.6.27 kernel. I
> think it is compiled etc. so i just renamed the vmlinux....2.6.27
> file into vmlinux and replace the vmlinux file from the FS-file
> package.
>
> Now i think every thing should work, but when i try to compile gem5 with
>
> scons /build/ALPHA_FS/
>
> a lot of seltests failed.
>
> For example this one:
>
> diff -ubrs -I '^command line:' -I '^M5 compiled ' -I '^M5 started '
> -I '^M5 executing on ' -I '^Simulation complete at' -I '^Listening
> for' -I 'listening for remote gdb' --exclude=stats.txt
> --exclude=outdiff
> tests/long/10.linux-boot/ref/alpha/linux/tsunami-o3
> build/ALPHA_FS/tests/debug/long/10.linux-boot/alpha/linux/tsunami-o3
> >
>
build/ALPHA_FS/tests/debug/long/10.linux-boot/alpha/linux/tsunami-o3/outdiff
> scons: *** Error 1
> ===== Output differences =====
> diff -ubrs -I '^command line:' -I '^M5 compiled ' -I '^M5 started '
> -I '^M5 executing on ' -I '^Simulation complete at' -I '^Listening
> for' -I 'listening for remote gdb' '--exclude=stats.txt'
> '--exclude=outdiff'
> tests/long/10.linux-boot/ref/alpha/linux/tsunami-o3/config.ini
>
build/ALPHA_FS/tests/debug/long/10.linux-boot/alpha/linux/tsunami-o3/config.ini
> ---
> tests/long/10.linux-boot/ref/alpha/linux/tsunami-o3/config.ini 2011-09-09
> 16:49:08.108057289 +0200
> +++
>
build/ALPHA_FS/tests/debug/long/10.linux-boot/alpha/linux/tsunami-o3/config.ini 2011-09-16
16:54:59.957208352
> +0200
> @@ -10,12 +10,12 @@
> children=bridge cpu disk0 disk2 intrctrl iobus iocache l2c membus
> physmem simple_disk terminal toL2Bus tsunami
> boot_cpu_frequency=500
>
> here the simerr file:
>
> warn: Sockets disabled, not accepting terminal connections
> warn: Sockets disabled, not accepting gdb connections
> warn: Prefetch instrutions is Alpha do not do anything
> warn: Prefetch instrutions is Alpha do not do anything
> hack: be nice to actually delete the event here
>
>
>
> summary:
> *****
>
build/ALPHA_FS/tests/fast/quick/80.netperf-stream/alpha/linux/twosys-tsunami-simple-atomic
> FAILED!
>
> *****
> build/ALPHA_FS/tests/opt/long/10.linux-boot/alpha/linux/tsunami-o3
> FAILED!
>
> *****
> build/ALPHA_FS/tests/opt/long/10.linux-boot/alpha/linux/tsunami-o3-dual
> FAILED!
>
> *****
>
build/ALPHA_FS/tests/opt/quick/10.linux-boot/alpha/linux/tsunami-simple-atomic
> FAILED!
>
> *****
>
build/ALPHA_FS/tests/opt/quick/10.linux-boot/alpha/linux/tsunami-simple-atomic-dual
> FAILED!
>
> *****
>
build/ALPHA_FS/tests/opt/quick/10.linux-boot/alpha/linux/tsunami-simple-timing
> FAILED!
>
> *****
>
build/ALPHA_FS/tests/opt/quick/10.linux-boot/alpha/linux/tsunami-simple-timing-dual
> FAILED!
>
> *****
>
build/ALPHA_FS/tests/opt/quick/80.netperf-stream/alpha/linux/twosys-tsunami-simple-atomic
> FAILED!
>
> It also the tests take very long time, some hours.
>
> I tried hard, so solve the problem by my self, but now i have no
> more ideas, how to solve the problem.
>
>
>
> By the way, when i try to boot linux i get this output:
>
> $ ./build/ALPHA_FS/m5.debug configs/example/fs.py
>
> gem5 Simulator System. http://gem5.org
> gem5 is copyrighted software; use the --copyright option for details.
>
> gem5 compiled Sep 16 2011 16:35:24
> gem5 started Sep 16 2011 18:56:54
> gem5 executing on hollerith
> command line: ./build/ALPHA_FS/m5.debug configs/example/fs.py
> Global frequency set at 1000000000000 ticks per second
> info: kernel located at: /gem5_rep/binaries/vmlinux
> Listening for system connection on port 3456
> 0: system.tsunami.io.rtc: Real-time clock set to Thu Jan 1
> 00:00:00 2009
> 0: system.remote_gdb.listener: listening for remote gdb #0 on port 7000
> **** REAL SIMULATION ****
> info: Entering event queue @ 0. Starting simulation...
> warn: Prefetch instrutions is Alpha do not do anything
> warn: Prefetch instrutions is Alpha do not do anything
>
> after that nothing happens for 20h+
>
> I think it is broken.
>
> Thank you very much for your help.
>
> Stephan Walter
>
>
> _______________________________________________
> 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
