I tested a little with the X86 architecture and SE mode, and results were
all correct. It seems to me that either the problem is in the FS mode, or
with the Alpha architecture.
--
Nilay
On Wed, 6 Jun 2012, GE ZHIGUO wrote:
Hi, Steve
Thanks for your advices which is very helpful!
I will have a try. Maybe, we can add some complicated tests into regression
test
and verify the correctness of the execution outputs.
Regards,
Zhiguo
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of
Steve Reinhardt
Sent: Wednesday, June 06, 2012 11:46 AM
To: gem5 Developer List
Subject: Re: [gem5-dev] GEM5 outputs are not right for PARSEC blackscholes
Thanks, this is very useful. I think the next step is to pick one of the
configurations that works for gem5-stable but not gem5-dev and see where it
breaks. You might as well pick the fastest running one, probably atomic
CPU with the classic memory system.
The 'hg bisect' command is designed for this task, but our change in
command-line arguments complicates that. (A good reason not to change
options lightly!) I suggest starting with the rev that changes the command
line arguments:
http://repo.gem5.org/gem5/rev/c3e7a961c727.
Depending on whether that rev is correct, then you can use 'hg bisect' to
go forward or back to find the bad changeset without having to deal with
the command lines changing on you.
Thanks again for helping track this down.
Steve
On Tue, Jun 5, 2012 at 8:31 PM, GE ZHIGUO <[email protected]> wrote:
I tried on gem5-stable and gem5-dev, and compare the results:
Ran parsec on GEM5 Full System simulation according to the following link:
http://www.cs.utexas.edu/~parsec_m5/
The execution command used in Full System linux:
./blackscholes 4 /parsec/install/inputs/blackscholes/in_16.txt
/parsec/install/inputs/blackscholes/prices_simdev.txt
GEM5-stable:
1. Normal FS (MI protocol ruby?):
build/ALPHA_FS/gem5.opt configs/example/ruby_fs.py --ruby --timing
Execution results: Correct.
2. Normal FS(classic memory?):
build/ALPHA_FS/gem5.opt configs/example/fs.py --timing
Execution results: Correct
3. ALPHA_FS_MOESI_CMP_token:
scons -j4 build/ALPHA_FS_MOESI_CMP_token/gem5.opt RUBY=true
build/ALPHA_FS_MOESI_CMP_token/gem5.opt configs/example/ruby_fs.py --ruby
--timing
Execution results: Wrong.
4. ALPHA_FS_MOESI_hammer:
scons -j4 build/ALPHA_FS_MOESI_hammer/gem5.opt RUBY=true
build/ALPHA_FS_MOESI_hammer/gem5.opt configs/example/ruby_fs.py --ruby
--timing
Execution results: Wrong
GEM5-dev:
1. Normal FS timing CPU:
scons -j4 build/ALPHA_FS/gem5.opt
build/ALPHA_FS/gem5.opt configs/example/fs.py --cpu-type=timing
Execution results: Wrong
2. Normal FS atomic CPU:
build/ALPHA_FS/gem5.opt configs/example/fs.py --cpu-type=atomic
Execution results: Wrong
3. Normal FS O3 CPU:
build/ALPHA_FS/gem5.opt configs/example/fs.py --cpu-type=detailed --caches
Execution results: Wrong
4. ALPHA_FS_MOESI_hammer:
scons -j4 build/ALPHA_FS_MOESI_hammer/gem5.opt RUBY=true
build/ALPHA_FS_MOESI_hammer/gem5.opt configs/example/ruby_fs.py --ruby
--cpu-type=timing
Execution results: Wrong
Conclusions:
1. GEM5-dev introduces bugs. The execution results for both class and ruby
memory are not correct.
2. Ruby does not work.
Interestingly, all the wrong results are the same.
Wrong execution results:
16
3.762516736984252930
-0.188307821750640869
2.030249118804931641
6.907309055328369141
0.930330634117126465
10.893478393554687500
14.152964591979980469
19.959602355957031250
1.463889837265014648
4.625942230224609375
8.334403991699218750
-0.032770160585641861
-0.083193600177764893
3.999399185180664062
10.102209091186523438
12.887604713439941406
Correct execution results:
16
4.759420394897460938
0.808597564697265625
3.714603424072265625
8.591663360595703125
2.133378982543945312
10.895606994628906250
14.421577453613281250
18.630859375000000000
1.814987182617187500
5.850273132324218750
10.308151245117187500
0.003523707389831543
1.140726089477539062
4.216762542724609375
11.135246276855468750
16.092636108398437500
-----Original Message-----
From: [email protected] [mailto:[email protected]] On
Behalf Of GE ZHIGUO
Sent: Wednesday, June 06, 2012 9:20 AM
To: gem5 Developer List
Subject: Re: [gem5-dev] GEM5 outputs are not right for PARSEC blackscholes
I tried yesterday, Ruby does not work for gem-stable version either. I
Only tried timing CPU, and
atomic and O3 CPU are not tried as gem5 status matrix shows these two CPU
do not support ruby well.
whether anyone had checked whether the execution results are correct
when ruby was enabled?
I will try to see whether ROI(region of interest) causes the issues.
-----Original Message-----
From: [email protected] [mailto:[email protected]] On
Behalf Of Ali Saidi
Sent: Tuesday, June 05, 2012 11:14 PM
To: gem5 Developer List
Subject: Re: [gem5-dev] GEM5 outputs are not right for PARSEC blackscholes
Yes, please bisect the repository and let us know which changeset is
causing issues. Do you have this problem with an atomic cpu, timing, o3?
Thanks,
Ali
On 05.06.2012 08:34, Mahmood Naderan wrote:
You
may try a binary search in changesets to find the problematic revision
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
