Hi Erwin,
I'm working on an EU-funded research project building a humanoid robot (see www.eccerobot.org).
We're currently intending using Bullet to implement a physics-based model of the robot which can then be used to develop, or form an active part of, control strategies.
However, we're going to need this complex model to run as fast as possible - certainly greater than real-time, so we are very keen to employ GPU acceleration.
The questions I have are:-
1. Can you give us some idea of how far off your Bullet 3.x OpenCL release is? i.e. weeks, months or years?
2. What areas of the engine will it accelerate? From what I can gather it will be primarily the collision engine and the constraints solver - which are the current performance bottlenecks anyway.
3. I'm running normal Bullet on Ubuntu 9.10. I have a NVidia GPU that runs CUDA code fine. I've been trying to test the OpenCL branch in svn. However, if you include the demos in the cmake then a lot of them don't compile for me , which of them do I need to run to demonstrate OpenCL accelerated Bullet in action?
Many Thanks
Alan Diamond
University of Sussex, UK