Research And Application Of GPU Based Tessellation

Thanks to the development of modern Graphics hardware，The graphics of3D videogame has significantly increased in quality and complexity during the last decade Especiallywhen programmable units are introduced into video card，3D game can now simulate muchriching world than before in realtime, with GPU shading technology like pixel lighting，shadowmap，soft particle，even none photorealistic effectHowever，realistic shading itself is not enough to create convincing representations ofrich worlds on the screen.For performance reason and memory storage consideration,lowresolution mesh are often used in game.But highly detail object is hard to represent by low resmesh.In fact, some shading technique like normal mapping are used to create the effect that atotally flat surface looks bumpy.But this illusion easily disappear when the angle betweenview vector and surface normal become larger. Advanced titles such as Crytek s Crysis arecreating complex surface effects via techniques such as parallax occlusion mapping,But thistechnique require lots of sampling operation in pixel shader，when sample count is low，aliasartifact will appear.Further more, low-resolution geometric representations always has acoarse silhouettes， this phenomenon can not be optimize with any shadingtechnology.Obviously, The need for true geometric detail has not completely gone away.A simple solution springs to mind–why not just increase the polygon count of theobjects in the game? There are at least three distinct costs associated with explicit geometricrepresentation.First, Meshes are an inherently expensive representation. We need to store afair amount of data pervertex, thus increasing the pressure on already-scarce memory. and puta strain on general memory bandwidth.Second, the animation for skinned mesh are oftenperform in vertex shader, when vertex count become large，this stage would be bottleneck dueto the expensive calculation.Third, large meshes display poor utilization of GPU vertexcaches, which will also impact performance.An alternative solution is store only low-res mesh in memory while tessellate itdynamicaly.However, if tessellation perform in cpu, we need to put the high-res mesh in gpuinevitablly，which cost a lot of bus bandwidth Fortunately，with the Directx11and Opengl4.0,we can do every tessellation process fully in GPU.This is the motive of this paper.