lock-to-one-core #f does not work on windows

Bug #307646 reported by Treeform
2
Affects Status Importance Assigned to Milestone
Panda3D
Fix Released
Undecided
Unassigned
1.6.x
Fix Released
Undecided
Unassigned

Bug Description

Running a program with lock-to-one-core #f does not work on windows xp sp2
It still sets affinity to a single core and effectively runs on a single core.
Versions 1.5.0 (maybe earlier) through 1.5.4
See http://panda3d.org/phpbb2/viewtopic.php?t=2859&highlight=affinity

while we are on the subject.
It would be cool if panda3d picks a core at random then starting 2 panda instances would use different cores at least.

Revision history for this message
rdb (rdb) wrote :

Hmm, do you even have threading enabled?

Revision history for this message
Treeform (starplant) wrote : Re: [Bug 307646] Re: lock-to-one-core #f does not work on windows

No, this has nothing to do with threading but with affinity. I should
be able to run a single thread on multiple cores or 4 instances of
panda on all 4 cores. Right now the 4 instances of panda3d all fight
for one core.

On Sat, Dec 27, 2008 at 11:07 AM, pro-rsoft <email address hidden> wrote:
> Hmm, do you even have threading enabled?
>
> --
> lock-to-one-core #f does not work on windows
> https://bugs.launchpad.net/bugs/307646
> You received this bug notification because you are a direct subscriber
> of the bug.
>

Revision history for this message
rdb (rdb) wrote :

Hmm, I can reproduce this bug with ppremake 1.6.0, on Ubuntu Intrepid 64b, with threading disabled, and the config vars correctly set. (I tried both lock-to-one-core and lock-to-one-cpu, which one is the correct one?)
The system monitor shows just 50% of my processor is in use.

Marked as Confirmed.

Changed in panda3d:
status: New → Confirmed
Revision history for this message
Treeform (starplant) wrote :

I don't think its a Linux bug. I think windows specific because only
on windows there is a problem with the clocking being different for
cores. It will show it to be 50% use that's normal. Look at program
affinity does it use only a single core?

Better experiment start 2 copies of your app. Do they use 100% of cpu
(one to each core) or 50% of cpu ( 25% of cpu each because they are
stuck to one core)

On Sat, Dec 27, 2008 at 11:55 PM, pro-rsoft <email address hidden> wrote:
> Hmm, I can reproduce this bug with ppremake 1.6.0, on Ubuntu Intrepid 64b, with threading disabled, and the config vars correctly set. (I tried both lock-to-one-core and lock-to-one-cpu, which one is the correct one?)
> The system monitor shows just 50% of my processor is in use.
>
> Marked as Confirmed.
>
> ** Changed in: panda3d
> Status: New => Confirmed
>
> --
> lock-to-one-core #f does not work on windows
> https://bugs.launchpad.net/bugs/307646
> You received this bug notification because you are a direct subscriber
> of the bug.
>

Revision history for this message
rdb (rdb) wrote :

When I run two copies of pview, cpu usage is 100%, because each pview runs on a different core. When I run just one copy of pview, cpu usage is 50%.

Revision history for this message
Treeform (starplant) wrote :

See then you don't have bug. If i run 2 pviews, they use 12% and 12%
respectively because i have 4 cores and they are using only one. This
is the bug.

On Sun, Dec 28, 2008 at 2:10 AM, pro-rsoft <email address hidden> wrote:
> When I run two copies of pview, cpu usage is 100%, because each pview
> runs on a different core. When I run just one copy of pview, cpu usage
> is 50%.
>
> --
> lock-to-one-core #f does not work on windows
> https://bugs.launchpad.net/bugs/307646
> You received this bug notification because you are a direct subscriber
> of the bug.
>

Revision history for this message
David Rose (droklaunchpad) wrote :

This was caused by another config variable, client-cpu-affinity, which was inadvertently defaulting to on. This has been fixed as of 1.6.

Changed in panda3d:
status: Confirmed → Fix Released
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers