This paper proposes a novel strategy based on fragment meshes of Shiue et al. for GPU rendering of compos- ite subdivision surfaces. Two enumeration systems are established to label the primitives of each fragment mes...This paper proposes a novel strategy based on fragment meshes of Shiue et al. for GPU rendering of compos- ite subdivision surfaces. Two enumeration systems are established to label the primitives of each fragment mesh. A sector-layer-index enumeration system is responsible for retrieving proximities for subdivision masks while a sec- tor-index enumeration system designates a 2D texture buffer in GPU. Recurring to the free conversion between them, our approach may get rid of lookup tables that are designed to record subdivision masks. In addition, relatively small composite subdivision masks make it easy to develop automatically retrieving techniques. Finally, as center vertices are often irregular, their computation is related to an average with alterable number of items. Considering that variable loop is not efficient in GPU, we evaluate the center vertex of each fragment mesh using the linear combination of its level 0, level 1 and limit positions instead of averaging schemes. Experiments demonstrate that our approach generally outper- forms that of Shiue et al. in FPS by a long way.展开更多
基金supported by Natural Science Foundation of China (60973084)Fundamental Research Funds for the Central Universities (2012zz0063)
文摘This paper proposes a novel strategy based on fragment meshes of Shiue et al. for GPU rendering of compos- ite subdivision surfaces. Two enumeration systems are established to label the primitives of each fragment mesh. A sector-layer-index enumeration system is responsible for retrieving proximities for subdivision masks while a sec- tor-index enumeration system designates a 2D texture buffer in GPU. Recurring to the free conversion between them, our approach may get rid of lookup tables that are designed to record subdivision masks. In addition, relatively small composite subdivision masks make it easy to develop automatically retrieving techniques. Finally, as center vertices are often irregular, their computation is related to an average with alterable number of items. Considering that variable loop is not efficient in GPU, we evaluate the center vertex of each fragment mesh using the linear combination of its level 0, level 1 and limit positions instead of averaging schemes. Experiments demonstrate that our approach generally outper- forms that of Shiue et al. in FPS by a long way.
