Improving GPU-Speedup of Optimized Volume Rendering

Abstract

This paper presents a speedup improvement method for optimized volume rendering in GPU platforms. First, from a set of experiments, we found that the speedup of volume rendering optimized with transparent voxel skipping decreases with dependency on the complexity of target images. In order to evaluate the complexity of volume images, we developed a new algorithm, called EVIC. Next, we present another new algorithm, called RBDV, that reduces the branch divergence in transparent voxel skipping by factoring out structurally similar code from branch paths in GPU programs. We empirically proved that this RBDV algorithm increases the GPU-speedup of transparent voxel skipping at least by 14%, improving it from x17.5 upto x20.0 or more, on average, for complex target images.