Extreme SIMD: Optimized Collision Detection in Titanfall [video]

This is an amazing talk from three years ago that deserves more attention. The use of divide by zero alone is mind blowing (for most people).

"Come discover how the 'Titanfall' team made already optimized continuous collision detection code more than twice as fast! This talk will go deep into the CPU and geometric optimizations that made this possible. Topics include AABB tree construction, rapid cache-friendly traversal of a bounding-volume hierarchy, and swept capsule vs triangle mesh collision. This presentation will show how to perform these operations efficiently on modern x64 CPUs' SIMD pipelines, including bit manipulation and register packing for branchless programming."