Difference in number of threads using --benchmarks option

98 views
Skip to first unread message

Rohith R

unread,
Nov 1, 2016, 1:44:59 AM11/1/16
to Sniper simulator
Hi,

I ran blackscholes from parsec using the command 
./run-sniper -n16 --roi -senergystats  -p parsec-blackscholes -i small -c gainestown and got 16 threads.

 But when I run it using the command :
./run-sniper -n16 -senergystats --benchmarks=parsec-blackscholes-small-16 - c gainestown I get 17 threads. Why is there a difference when using both the options ?

Is it because of the fact that the threads in blackscholes are generated inside the ROI markers and while using the --benchmarks option the ROI markers are ignored (not totally) i.e. the applications are simulated from start to end completely ?

Regards,
Rohith R

Wim Heirman

unread,
Nov 1, 2016, 5:22:35 AM11/1/16
to snip...@googlegroups.com
Hi Rohith,

The reason is that the -n switch calculates a new core count parameter, such that the workload doesn't spawn more threads than the -n value; whereas --benchmarks passes the value directly to the workload. This happens in benchmarks/parsec/__init__.py:Program:get_nthreads: when using --benchmarks, nthreads_force is used so the value is used directly.

Regards,
Wim


--
--
--
You received this message because you are subscribed to the Google
Groups "Sniper simulator" group.
To post to this group, send email to snip...@googlegroups.com
To unsubscribe from this group, send email to
snipersim+unsubscribe@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/snipersim?hl=en

---
You received this message because you are subscribed to the Google Groups "Sniper simulator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to snipersim+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Rohith R

unread,
Dec 31, 2016, 4:50:12 AM12/31/16
to Sniper simulator
Hi Wim,

Is there any fix to this yet. Can I get the corr3ect number of threads while using the --benchmarks option by modifying the __init__.py script ?

I was looking at code and tried different things and could not get it to work. Any help would be appreciated here.

Regards,
Rohith R


On Tuesday, November 1, 2016 at 5:22:35 PM UTC+8, Wim Heirman wrote:
Hi Rohith,

The reason is that the -n switch calculates a new core count parameter, such that the workload doesn't spawn more threads than the -n value; whereas --benchmarks passes the value directly to the workload. This happens in benchmarks/parsec/__init__.py:Program:get_nthreads: when using --benchmarks, nthreads_force is used so the value is used directly.

Regards,
Wim

On 1 November 2016 at 06:44, Rohith R <rohi...@gmail.com> wrote:
Hi,

I ran blackscholes from parsec using the command 
./run-sniper -n16 --roi -senergystats  -p parsec-blackscholes -i small -c gainestown and got 16 threads.

 But when I run it using the command :
./run-sniper -n16 -senergystats --benchmarks=parsec-blackscholes-small-16 - c gainestown I get 17 threads. Why is there a difference when using both the options ?

Is it because of the fact that the threads in blackscholes are generated inside the ROI markers and while using the --benchmarks option the ROI markers are ignored (not totally) i.e. the applications are simulated from start to end completely ?

Regards,
Rohith R

--
--
--
You received this message because you are subscribed to the Google
Groups "Sniper simulator" group.
To post to this group, send email to snip...@googlegroups.com
To unsubscribe from this group, send email to

For more options, visit this group at
http://groups.google.com/group/snipersim?hl=en

---
You received this message because you are subscribed to the Google Groups "Sniper simulator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to snipersim+...@googlegroups.com.

Wim Heirman

unread,
Dec 31, 2016, 6:10:31 AM12/31/16
to snip...@googlegroups.com
Hi Rohith,

Whether or not to do the thread count correction is specified by the force_nthreads option. In run-sniper it's automatically added when using --benchmarks on line 287. I think removing it from there should be enough.

Regards,
Wim



For more options, visit this group at
http://groups.google.com/group/snipersim?hl=en

---
You received this message because you are subscribed to the Google Groups "Sniper simulator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to snipersim+unsubscribe@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages