Page 1 of 3
Bullet Optimisation
Posted: 11 Sep 2011, 14:11
by gus
Let's starts back where we finished last time:
First, here is the link to the previous discussion:
http://openmw.org/forum/viewtopic.php?f=13&t=129
So K1ll, could you do some testing for me? More specifically, could you go in libs/openengine/bullet/physic.cpp and look at line 285 :
Code: Select all
dynamicsWorld->stepSimulation(deltaT,1,1/50.);
and try out some different values? For exemple:
Code: Select all
dynamicsWorld->stepSimulation(deltaT,1,1/30.);
Code: Select all
dynamicsWorld->stepSimulation(deltaT,1,1/20.);
Code: Select all
dynamicsWorld->stepSimulation(deltaT,1,1/10.);
Code: Select all
dynamicsWorld->stepSimulation(deltaT,1,1.);
Code: Select all
dynamicsWorld->stepSimulation(deltaT,1,5.);
And report the FPS?
Could you also do it for this?
Code: Select all
dynamicsWorld->stepSimulation(deltaT,10,1/10.);
Code: Select all
dynamicsWorld->stepSimulation(deltaT,0,1/10.);
Re: Bullet Optimisation
Posted: 11 Sep 2011, 23:13
by K1ll
Sorry for replying so late, but i was a bit busy this weekend.
I'll test those changes tomorrow at least on my old laptop mentioned in the old thread and my ubuntu 10.10 64 bit desktop.
Re: Bullet Optimisation
Posted: 12 Sep 2011, 16:16
by gus
Thanks!
BTW,5 hours is pretty short for me
Re: Bullet Optimisation
Posted: 12 Sep 2011, 16:49
by K1ll
Well i have run tests with all the values you suggested. These are the results:
Parameter Value(s) | Interior fps (in Beshara) | Exterior fps (in Vivec)
Old Laptop:
1/50. | 28 | < 0
1/30. | 28-29 | < 0
1/20. | 51-53 | < 0
1/10. | (75) 80-85 | < 0
1. | 112-116 | < 0
5. | 113-117 | 0-11
10., 1/50. | 3 | < 0
0. 1/50. | 28-29 | < 0
Dekstop:
1/50. | 65-66 | 1
1/30. | 106-107 | 1
1/20. |124 | 1
(1/10. | 143 | 1
1.| 158-159 | 7-8
5. | 159-161 | 5-20
10., 1/50 | 63-66 | 1
0., 1/50 | 57-58 | 1
Re: Bullet Optimisation
Posted: 12 Sep 2011, 20:33
by gus
Could you do the same bench (just do few of them, like 1/50., 1/10. and 1.) with line 346 to 369 commented? (you can use /* */ to comment things)
And can i also have your desktop config?
Re: Bullet Optimisation
Posted: 13 Sep 2011, 12:50
by K1ll
Here are the tests with the changes applied:
Old Laptop
1/50. | 28-29 | 0
1/10. | 85-86 | 0
1. | 110-113 | 0
5. | 111-115 | 4-12
Dekstop
1/50. | 72-75 | 1
1/10. | 143-144 | 1
1. | 159-160 | 8
5. | 162-163 | 8-22
Not much of a difference though.
The specs of my Desktop:
Ubuntu Maverick Meerkat 10.10 64 bit
AMD Athlon 64 X2 Dual Core Processor 6000+
8 GB DDR2 Ram
Nvidia GeForce 9800 GTX+ with 512 MB video ram
Re: Bullet Optimisation
Posted: 13 Sep 2011, 14:50
by gus
Would it be possible to have a test on windows on the same machines? This would be very useful, as if it run correctly on windows, it might be a bug in bullet and not in OpenMW.
PS:a very stupid question: you compiled in release mode right?
Re: Bullet Optimisation
Posted: 13 Sep 2011, 16:55
by K1ll
I don't have windows on the old laptop (it runs way to slow on it) but i can test it with win xp and win 7 on my desktop if you could compile the tests and upload the binaries.
I'm pretty sure that i don't use debug binaries atleast i didn't specify it explicitly.
Re: Bullet Optimisation
Posted: 14 Sep 2011, 21:05
by gus
Here is a windows version:
http://www.megaupload.com/?d=8I70P6V5
Edit: BTW Zini, you can merge my bullet_bug branch in the main branch. There is a performance boost in this branch.
Re: Bullet Optimisation
Posted: 18 Sep 2011, 18:17
by K1ll
I've tested on Windows XP and it has similar slowdowns they are just not as noticeable since sadly openmw seems to run a lot faster on windows in general.
In Beshara i had ca 250 fps and in Vivec it still managed to get 70+ fps. Without pyhsics i had nearly 500 fps in Beshara and about 200 fps in Vivec.