Real-Time Physics Simulation Forum

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-08-14T15:15:09+00:00

is it possible to get a changes summary of the latest 2.75 RC here ? (2.75 RC7 atm)

Statistics: Posted by Cleanrock — Fri Aug 14, 2009 3:15 pm

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-11T19:16:50+00:00

Why was split impulse removed?

And when was it removed? I'm using 2.73, was it in that?

I ask because we are seeing, every once in a while, some very strange output from Bullet (position data wise), and I'm wondering if it could be from a bug in the split impulse solver?

Too be specific we are seeing NAN as position output (Wii/PPC)

We are right on the cusp of release, so we don't want to upgrade any libs if we don't have to, but if there is a fixed split impulse solver in this version, we might have to.

Statistics: Posted by Wavesonics — Thu Jun 11, 2009 7:16 pm

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-11T14:18:18+00:00

Thanks, can you try out Bullet 2.75 RC2?

Yes, it works fine now! Thank you.

Statistics: Posted by ola — Thu Jun 11, 2009 2:18 pm

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-11T14:09:08+00:00

EDIT: Maybe it's sufficient to add something like this to btDiscreteDynamicsWorld::addRigidBody? "else setActivationState(ISLAND_SLEEPING);" That fixed the issues I had here in my code.

Thanks, can you try out Bullet 2.75 RC2?
Erwin

Statistics: Posted by Erwin Coumans — Thu Jun 11, 2009 2:09 pm

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-11T11:01:37+00:00

I just tested the RC1 (svn 1691), and I think I might have a problem with static rigid bodies. We depend on switching the rigid bodies from dynamic to static and back (by removing, re-adding with or without mass).

Previous behaviour for these was that they are added to the world and show up as active for a second or so, before turning inactive (they return body->getActivationState() == ISLAND_SLEEPING). Now, they appear to remain active at all times, returning body->getActivationState() == ACTIVE_TAG.

I don't know if the only problem is that they return ACTIVE_TAG instead of ISLAND_SLEEPING, or if this then also influences the collision detection performance? At least, there might be code that relies on this flag, which will now break because of this change. I'm looking through my own code for this right now...

I also saw that now btDiscreteDynamicsWorld::setGravity and clearForces only applies to the dynamic rigid bodies. This could have broken some functionality in our case, so please make a note of this in the release notes

Apart from that, the improvement in only updating the moving rigid body transforms is great

Best regards,
Ola

EDIT:
Maybe it's sufficient to add something like this to btDiscreteDynamicsWorld::addRigidBody? "else setActivationState(ISLAND_SLEEPING);" That fixed the issues I had here in my code.

Code:

voidbtDiscreteDynamicsWorld::addRigidBody(btRigidBody* body){if (!body->isStaticOrKinematicObject()){body->setGravity(m_gravity);}if (body->getCollisionShape()){if (!body->isStaticObject()){m_nonStaticRigidBodies.push_back(body);}            else                body->setActivationState(ISLAND_SLEEPING);bool isDynamic = !(body->isStaticObject() || body->isKinematicObject());short collisionFilterGroup = isDynamic? short(btBroadphaseProxy::DefaultFilter) : short(btBroadphaseProxy::StaticFilter);short collisionFilterMask = isDynamic? short(btBroadphaseProxy::AllFilter) : short(btBroadphaseProxy::AllFilter ^ btBroadphaseProxy::StaticFilter);addCollisionObject(body,collisionFilterGroup,collisionFilterMask);}}

Statistics: Posted by ola — Thu Jun 11, 2009 11:01 am

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-11T07:36:32+00:00

Bullet 2.75 RC1 has been uploaded, based on latest SVN trunk revision 1691.
I'm trying to compile the sph project and it's looking for the glee_2008d.lib file. Any idea what this is, or rather, where it is ?
Can you please try RC1? glee*.lib should not be needed anymore, it includes glee.c now.

Thanks Erwin, it compiles now

Cheers.

Statistics: Posted by JamesH — Thu Jun 11, 2009 7:36 am

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-11T01:42:18+00:00

Bullet 2.75 RC1 has been uploaded, based on latest SVN trunk revision 1691.

I guess I am getting a bit annoying with my stupid feature request, but any idea when split impulse will be there again ?

Can you test Bullet 2.75 RC1, it adds the split impulse option.

(virtual definition of addCollisionObject / removeCollisionObject in btCollisionWorld.h)

Added in RC1.

I'm trying to compile the sph project and it's looking for the glee_2008d.lib file. Any idea what this is, or rather, where it is ?

Can you please try RC1? glee*.lib should not be needed anymore, it includes glee.c now.

Atleast, I did myself simple modification to btDiscreteDynamicsWorld

Same simple modification has been applied, can you please test RC1?

Thanks for all the feedback!
Erwin

Statistics: Posted by Erwin Coumans — Thu Jun 11, 2009 1:42 am

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-09T13:52:36+00:00

Just a small reminder for this tiny fix
http://www.bulletphysics.com/Bullet/php ... f=9&t=3599

(virtual definition of addCollisionObject / removeCollisionObject in btCollisionWorld.h)

Best regards,
Ola

Statistics: Posted by ola — Tue Jun 09, 2009 1:52 pm

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-06-09T07:19:51+00:00

I'm trying to compile the sph project and it's looking for the glee_2008d.lib file. Any idea what this is, or rather, where it is ?

Many thanks

Statistics: Posted by JamesH — Tue Jun 09, 2009 7:19 am

Re: Bullet 2.75 beta1: GPU, SPH fluids preview, new constraints

2009-05-30T12:34:25+00:00

Just another reminder - I don't see "active objects set" in 2.75beta.
It still iterates m_collisionObjects many times each simulation step.
Atleast, I did myself simple modification to btDiscreteDynamicsWorld
that I have separate container for rigid bodies thus many loops with
such construction:
for (int i=0;i{
btCollisionObject* colObj = m_collisionObjects;
btRigidBody* body = btRigidBody::upcast(colObj);
if (body)

are working with much smaller set of data now. Until that, it made
Bullet unusable for large world with thousands of static objects and
only a few rigid bodies.

best regards, Fido

Statistics: Posted by fido — Sat May 30, 2009 12:34 pm