Hey let's switch away from bullet to anything else (discussion)

Everything about development and the OpenMW source code.
User avatar
akortunov
Posts: 605
Joined: 13 Mar 2017, 13:49
Location: Samara, Russian Federation

Re: Hey let's switch away from bullet to anything else

Post by akortunov » 25 Nov 2018, 08:52

Can we contact with Bullet team and ask how we can optimize physics?
Basically, we have a world with triangle-based collision shapes and dynamic actors with capsule or AABB shapes.
Most of time for now is spent in convexSweepTest's, which we perform one by one in the main thread. Is there a better approach?
How do other games use Bullet?

FYI: a Bullet developer lunkhound made a good progress with multithreading, maybe there could be something useful for us.

Also I found an interesting thread, where TS managed to run parallel convexSweepTests by building Bullet with thread locks.

User avatar
psi29a
Posts: 4618
Joined: 29 Sep 2011, 10:13
Location: Belgium
Gitlab profile: https://gitlab.com/psi29a/
Contact:

Re: Hey let's switch away from bullet to anything else

Post by psi29a » 25 Nov 2018, 10:40

Sure.

There is no reason not to contact people that work on libraries we use.

User avatar
akortunov
Posts: 605
Joined: 13 Mar 2017, 13:49
Location: Samara, Russian Federation

Re: Hey let's switch away from bullet to anything else

Post by akortunov » 26 Nov 2018, 06:12


User avatar
psi29a
Posts: 4618
Joined: 29 Sep 2011, 10:13
Location: Belgium
Gitlab profile: https://gitlab.com/psi29a/
Contact:

Re: Hey let's switch away from bullet to anything else

Post by psi29a » 26 Nov 2018, 09:54

akortunov wrote:
26 Nov 2018, 06:12
https://github.com/bulletphysics/bullet3/issues/2004

I hope they will response.
Great! :)

There was a bug in bullet long ago when it was still being hosted on google code and I asked everyone in the forums to 'vote' on the issue. It became so popular that it became the first thing on their priority list to fix. Pity that Github and Gitlab don't have this kind of popularity mechanism so that it gets visibility.

I guess if people are interested, they can just +1 it.

Chris
Posts: 1558
Joined: 04 Sep 2011, 08:33

Re: Hey let's switch away from bullet to anything else

Post by Chris » 04 Dec 2018, 05:00

Just to throw it out there, nVidia just open sourced PhysX, under the simple BSD-3 license. Should be usable by GPL apps, though I can't say anything about the API or how easy it is to use.

User avatar
psi29a
Posts: 4618
Joined: 29 Sep 2011, 10:13
Location: Belgium
Gitlab profile: https://gitlab.com/psi29a/
Contact:

Re: Hey let's switch away from bullet to anything else

Post by psi29a » 04 Dec 2018, 09:19

That is wonderful news! :)

User avatar
gus
Posts: 390
Joined: 11 Aug 2011, 15:41

Re: Hey let's switch away from bullet to anything else

Post by gus » 04 Dec 2018, 12:00

I've used it quite some time ago. Can't comment about the performances, but the documentation was rock solid, which is what Bullet lacks. It sill looks very good: https://docs.nvidia.com/gameworks/conte ... Index.html.

Might be a good choice to switch to IMO, just for the quality of the doc.

User avatar
Br0ken
Posts: 241
Joined: 02 Apr 2012, 05:54
Location: Siberia

NVIDIA open sources PhysX

Post by Br0ken » 04 Dec 2018, 19:58

NVIDIA has announced that simulation engine PhysX is to go open source under the simple BSD-3 license for desktop and mobile applications. The EULA remains unchanged for applications that run on XBox One, Playstation 4, and Switch.

Read the announcement on NVIDIA’s blog and access the source code on Githib.
Can it be useful for OpenMW? ;)

User avatar
lysol
Posts: 1307
Joined: 26 Mar 2013, 01:48
Location: Sweden

Re: NVIDIA open sources PhysX

Post by lysol » 04 Dec 2018, 20:01

Hmm, maybe we should try to find some kind of "official" discussion thread about this subject, since this has been discussed at several channels at the discord server, at least once in the irc server, and at least here on the forums.

Most interesting discussion was probably at the irc server. A copypasta of it:
[irc] Ananace: So apparently PhysX has been relicensed under the BSD 3-clause now, perhaps it might be worth taking a short look at it as well for a possible solution for the physics performance issues
[irc] dmbaturin: Does it work on non-nvidia cards now?
[irc] jvoisin: https://news.developer.nvidia.com/annou ... cs-engine/
[irc] jvoisin: the sdk is under BSD3
[irc] johnnyhostile: i was gonna say, that cant be true haha
[irc] dmbaturin: shakes off liquid nitrogen left from his cryostasis chamber
[irc] johnnyhostile: nvidia does not usually play like that
[irc] serpn: Hahah, that's incredible
[irc] serpn: so very nvidia
[irc] jvoisin: everybody on the internet is singing that nvidia is so cool for open-saucing physx
[irc] jvoisin: sigh
[irc] johnnyhostile: "mission accomplished"
[irc] serpn: but we can edit the headers!?!?
[irc] serpn: I guess in a way it means that its more-possible for developers (AAA) to wrap and emulate/reimplement.... but ye useless for 'open source'
[irc] johnnyhostile: just another PR move, nothing to see here.
[irc] serpn: Not that anyone has cared about that shit since Google vs. Oracle, buuuut ye
[irc] serpn: nah, knowing guys who have run into trouble hacking middleware/sdk stuff for commercial projects, there's likely a liiiiiittttttle bit of good news there, but they have shined that turd up for PR for sure.
[irc] johnnyhostile: sure, as you indicated its likely going to affect a very specific audience
[irc] serpn: The initial physx stuff, before nvidia purchased it was pretty interesting
[irc] serpn: I remember playing with the demos from NovodeX? or whoever it was
[irc] serpn: and then two months after the purchase the same demos ran at like half speed, magically.
[irc] serpn: but ye fuck nv toot toot :)
[irc] jvoisin: there was a game to showcase physx that was impressive
[irc] jvoisin: fighting by throwing a fuckton of objects at each other
[irc] johnnyhostile: i am inclined to agree with you re: 'fuck nv toot toot' but i am a hippocrite and use an nvidia GPU and the closed driver :X
[irc] serpn: same johnnyhostile
[irc] johnnyhostile: muh performance
[irc] jvoisin: mh, the video on nvidia's physx page is still looking pathetic tbf :/
[irc] serpn: but I bought a non-Ti 1050, just because of that :)
[irc] jvoisin: shuttering everywhere :/
[irc] johnnyhostile: like it or not, nvidia cards are 1st-class for Proton
[irc] serpn: hah, it was NovodeX, I even got that capitalization correct
[irc] johnnyhostile: and i am loving that angle.
[irc] jvoisin: also, why is the video showcasing machine-learning bullshit?
[irc] johnnyhostile: my guess would be more buzzword/PR points
[irc] jvoisin: sigh
[irc] serpn: RTX: on
AnyOldName3: Bear in mind that PhysX is primarily a CPU-based physics engine. Unless it's only used for one flashy effect and everything else is done using Havok or Bullet (e.g. Fallout 4), most games only use the CPU-only component. In fact, there have been no hardware-accelerated PhysX games yet this year, and it's nearly over.
[irc] jvoisin: wow
[irc] jvoisin: I thought that it was heavily gpu-based
[irc] jvoisin: since physx used to sell graphic cards "extensions" iirc
[irc] jvoisin: https://en.wikipedia.org/wiki/Ageia
[irc] serpn: ye its been cpu centric from the start, the initial novodex sdk was all about cpu. nvidia pessimised the compilation of the library later to use more X87 instructions, and crippled some of the SSE stuff involved after buying if from Ageia.
[irc] serpn: ps. speak to nv staff if ever you get the chance, their technical guys are only there for the money.
[irc] serpn: sort of waiting for Oracle to buy them at this point.
AnyOldName3: How aware of AdoredTV are people here? He's got some pretty good videos on the history of Nvidia's anti-competitive practices, but they're pretty long as there's so much to cover and the level of detail which he goes into.
[irc] serpn: interesting, will take a look
[irc] johnnyhostile: nice, thanks anyoldname3
Normal mapped texture replacers, exclusive for OpenMW:
My Nexus page

User avatar
psi29a
Posts: 4618
Joined: 29 Sep 2011, 10:13
Location: Belgium
Gitlab profile: https://gitlab.com/psi29a/
Contact:

Re: Hey let's switch away from bullet to anything else

Post by psi29a » 04 Dec 2018, 20:09

Merged the threads.

Post Reply