Fast Construction of SAH-based Bounding Interval Hierarchies using Dynamic Parallelism

We present a fast and efficient algorithm for the construction of a SAH-based
BIH on the GPU. Our approach uses a novel asynchronous processing scheme which launches
kernels for the necessary subtasks directly on the GPU. This avoids the communication over-
head between CPU and GPU and optimizes the GPU’s workload. The SAH is employed for
all hierarchy levels which results in very efficient BIHs. Our results show that our algorithm
processes hundreds of thousands primitives at interactive frame rates. This enables interactive ray tracing of dynamic scenes even with changing geometry as is necessary for skeletal
animation or adaptive tessellation.