Session Name: | GPU Rigid Body Simulation |
Speaker(s): | Erwin Coumans |
Company Name(s): | Bullet Physics |
Track / Format: | Programming |
Overview: | General purpose computing on GPU is becoming mainstream in gaming hardware such as consoles, mobile devices, and PCs. Code that effectively uses the GPU can increase performance and improve battery life. This lecture discusses how to efficiently implement collision detection, particles, and rigid body simulation executed on the GPU. Although OpenCL is taught, most information is applicable to CUDA and DirectCompute. For each stage in the particle and rigid body simulation pipeline, we will discuss the suitability of popular algorithms for GPU. This includes collision detection, contact generation and contact reduction for convex and concave triangle meshes, contact constraint setup, and solving. You can expect to simulate up to hundred thousand active rigid bodies with stable stacking in real-time on modern GPUs. Everything discussed in the lecture is implemented and available as open source under a permissive license. This means that attendees can experiment, make changes, or reuse code in their own games. |