SAH KD-tree Construction On GPU

As the revolutionary development of GPU parallel computing, especially the CU-DA tech, the General-Purpose computation on Graphics Processing Unit(GPGPU) has pla-yed a significant role in General-Purpose computation. Even several years ago, the GP-GPU just can handle some simple computation. Consequently, many time-consu-ming questions receive much concern and study from the view of parallel. Especially a lot of traditional computations of computer graphic, most of which are data-intensive problem and very suitable for parallel computing.This paper has discoursed learnedly on a fundamental problem of Computer Gra-phic that is how to designed high efficient and robust KD-Tree on GPU for ray-tracing rendering. Based on explicitly analysis of existing measures, author resents a novel al-gorithm to build KD-Tree on GPU using the purely SAH optimal function with non-d-iscrimination of Large/small nodes and solving the "perfect split" problem. This algor-ithm is adapted to modern GPU processing architecture and can be easy to coding. For ordinary size scene, this new algorithm can achieve interactive rate and generate phot-o-realistic rendering result.