Geometry Shader
- 1. Optimizations with Geometry Shaders Alessandro Rigazzi Data Analysis and Visualization Group May 27, 2008
- 2. Geometry shader: what and where Programmable stage of the rendering pipeline In Open GL: takes any primitive as input, can output: Points Line Strips Triangle Strips Basically: can add geometry which is not stored Vertex Shader Geometry Shader Fragment Shader
- 3. Some potential uses Mesh refinement Send few triangles to video memory and refine them in shader Frustum culling Discard triangles outside of rendered scene before fragment processing Level Of Detail Refine mesh according to distance from camera We merge these three ideas in one sample application
- 4. Sample application: first stage, refinement We draw spheres: for each sphere we just send two pyramids ( eight triangles ) to the video memory and we refine them three times in the geometry shader Input Geometry Output Geometry
- 5. Sample application: second stage, LOD We dynamically subdivide the spheres, depending on their distance to the camera, framerate is more than doubled (four levels of detail) Input Geometry Output Geometry
- 6. Sample application: third stage, frustum culling We do not process triangles which are completely outside of the frustum : the image is the same, but we get an 11x improvement of the framerate Input Geometry Output Geometry
- 7. Benchmark
- 8. Limitations Geometry shaders require Shader Model 4 , i.e., for Open GL two extensions are needed: GL_EXT_geometry_shader4 GL_EXT_gpu_shader4 Extensions supported in modern graphic cards (e.g. nVidia G80 or later) Limited number of primitive and vertices which can be emitted by the vertex shader (commonly, at most 1024 floating points components can be emitted)