[osg-users] Announcing osgWorks and osgBullet
pmartz at skew-matrix.com
Wed Nov 11 08:59:56 PST 2009
Alberto Luaces wrote:
> Hi Paul,
> Lovely work! I assume that for now, independent threading for the
> physics is handled only by the multi-threaded Bullet library, is this
> correct? I couldn't find in the source any place using the
> PhysicsThreading class, are you going to use it in the future?
> I couldn't be able to compile the library since Bullet disables the
> building of the BulletMultiThreaded library on 64 bit environments,
> being it a requirement for osgBullet.
Hi Alberto -- Thanks for trying osgBullet.
After the 2.74 release, the Bullet people made a series of incompatible
changes to BulletMultiThreaded. I've modified osgBullet to not use
BulletMultiThreaded if the version is greater than 2.74, but probably
failed to make the CMake system robust enough to handle the library's
absence. I'll add this to my to-do list.
BulletMultiThreaded contains some host-thread SPU classes to do
constraint solving in parallel. In my investigation, these classes
typically run slower than the non-threaded sequential solver.
osgBullet allows an application to run the physics simulation in a
separate host thread, so that OSG rendering isn't blocked by the cost of
the physics simulation. That separate thread can use either the
sequential solver or the host-thread solver; osgBullet doesn't impose
any restrictions on how the simulation is performed by Bullet.
I do have a few simple examples that show how to do this, but they use
proprietary client data. I'll add to my to-do list, either to rewrite
this in a non-proprietary way, or get permission from the client to open
source the data.
More information about the osg-users