Cost Distribution Prediction for Parallel Ray Tracing
Erik Reinhard, Arjan J. F. Kok, Alan G. Chalmers · Bristol Research (University of Bristol) · 1998
Realistic rendering algorithms such as ray tracing are computationally very expensive. The time complexity is determined by a multitude of factors such as image size, geometry, material properties and the quality of the required rendering. This paper presents an algorithm to estimate the cost of ray tracing a scene. Assuming an octree spatial subdivision, the cost per voxel is predicted. For parallel processing purposes, this cost distribution can then be used to calculate an initial data distribution, such that the workload per processor should be more or less equal. The method avoids intersection tests, because those may affect the short response times required for preprocessing stages.