Technical Report on the Shader Model 4.0 Architecture

<strong>Technical</strong> <strong>Report</strong> on the Shader Model 4.0 ArchitectureShiben Bhattacharjee ∗ Suryakant Patidar † Jag Mohan Singh ‡ P. J. Narayanan §Center for Visual Information Technology, IIIT HyderabadFigure 1: (a) Geometry generation with Geometry Shader, (b) Physics interactions using Stream Output, (c) terrain rendering with ArrayTextures, (d) Multi Light Shadows with layered renderingAbstractThe Direct3D10/SM4.0 system [?] is the 4 th generation programmablegraphics processing units (GPUs) architecture. The newpipeline introduces significant additions and changes to prior generationpipeline. We explore these new features and experimentto judge their performance. The main facilities introduced thatwe ponder upon are, Unified Architecture providing common featuresset for all programmable stages, Goemetry Shader which isa new programmable stage capable of generating additional primitives,Stream output with which primitve data can be streamed tomemory, Array textures and primitive level redirection to differentframebuffers through layered rendering. We analyze our implementationsand with experimentation, we draw conclusions on their efficientusage, some of their limitations and suggestions for futureversions.Keywords: Shader Model 4.0, Direct3D 10 System, programmablegraphics hardware1 IntroductionIn real-time 3D computer graphics, a rendering pipeline is meantto accept certain representation of a 3D scene as input and producea 2D raster image as output which can be shown on a displaydevice. The various pipeline stages are mentioned in Figure ??.Earlier, these stages were implemented as fixed functions on hardwarefor graphics acceleration. Later, some of these stages weremade programmable instead of being fixed function namely processingof vertices and fragment. Since all vertices and fragmentscan be thought of as independent, they can be processed parallely∗ e-mail:shiben@research.iiit.ac.in† e-mail:skp@research.iiit.ac.in‡ email:jagmohan@research.iiit.ac.in§ e-mail:pjn@iiit.ac.inon the graphics processing unit. Small programs could be writtenwhich processed all vertices (or fragments) parallely and werecalled shaders and the programming model was called the shadermodel. The history of graphics hardware and shader model is discussedin Section 2.The Shader Model 4.0 makes giant leaps considering a versionchange, as it involves significant changes to the rendering pipeline.Earlier vertex shaders and fragment shader, being at differentstages, had different requirements. Fragments shaders had quickaccess to textures and vertex shaders either didn’t have or had limitedaccess to textures. Shader Model 4.0 keeps all the shaders atthe same level calling it a unified architecture, because of whicheven vertex shader (and geometry shader) have fast access to textures.This feature can be used to draw objects whose geometryinformation is stored as images, a very good example being constructingobjects with geoemtry images [?]. We discuss usage ofunified architecture in Section 3.Shader Model 4.0 introduces a new programmable stage called thegeometry shader which gives the capability of generating more verticesand primitives. This saves the CPU to GPU bandwidth as less3D data can be sent using programming tricks to the pipeline andthey can be generated on the GPU itself. We stress test the capabilitiesand verify the importance of this new programmable stage(Section 4). We develope techniques and applications involving geometryshader and our analysis will be helpful in clever usage ofgeometry shader.In past, game developers complained about slow state change becauseof which they avoided using multiple textures. Instead theymapped textures on various objects with the help of a single texture,texture atlas [?]. Shader Model 4.0 introduces array textures and 3Dtextures. Various images can be loaded as layers in array texture (oras a 3D texture) and the shader can choose between these layers asthe whole array texture is referenced by a single texture ID. Terrainrendering finds itself as a good application of array textures whichwe explain in detail in Section 5.Along with the introduction of array textures, another new attractivefeature is layered rendering which seems an extension of oldermodel’s one of the feature, MRT (Multiple Rendering Targets) [?].Where with MRT, fragments in the fragment shader can be redirectedto different color attachments of the framebuffer; with layeredrendering, it is possible in the geometry shader to redirectprimitives to different layers of an array texture which are attached

to the frame buffer object. Independent rasterization for all the differentlayers will be carried out. We find that multi-perspectiverendering and multiple light shadows are good applications of layeredrendering and are discussed in Section 6. Real-time Cube-maprendering is one more good application which has been discussedin ??.The new rendering pipeline also features streaming data out in themiddle of the pipeline to the memory. This is termed as streamoutput by the Direct3D 10 system and transform feedback by theOpenGL 2.1 system. This gives the facility of not going throughthe whole pipeline to process stream data which is not to be rendered.Objects can be rendered as well as some information canbe recorded in the same pipeline. We implemented simple physicstransformations and deformable models which harnesses the transformfeedback’s facilities; they are described in Section 7.We conclude the technical report with results and screenshots. Wealso some future work and some suggestions of improvement onthe new shader model in Section 8.2 History1. Mention of voodoo cards, 3dfx api 2. mention of nvidia and atiemergence 3. directx and opengl 4. introduction of shader model 5.evolution of shader model untill 4.03 Unified Architecture4 Geometry Shader4.1 Geometry generation and deletionKARTCH, D. 2000. Efficient Rendering and Compression for Full-Parallax Computer-Generated Holographic Stereograms. PhDthesis, Cornell University.LANDIS, H., 2002. Global illumination in production. ACM SIG-GRAPH 2002 Course #16 Notes, July.LEVOY, M., PULLI, K., CURLESS, B., RUSINKIEWICZ, S.,KOLLER, D., PEREIRA, L., GINZTON, M., ANDERSON, S.,DAVIS, J., GINSBERG, J., SHADE, J., AND FULK, D. 2000.The digital michelangelo project. In Proceedings of SIGGRAPH2000, ACM Press / ACM SIGGRAPH, New York, K. Akeley,Ed., Computer Graphics Proceedings, Annual Conference Series,ACM, 131–144.PARK, S. W., LINSEN, L., KREYLOS, O., OWENS, J. D., ANDHAMANN, B. 2006. Discrete sibson interpolation. IEEE Transactionson Visualization and Computer Graphics 12, 2 (Mar./Apr.), 243–253.PARKE, F. I., AND WATERS, K. 1996. Computer Facial Animation.A. K. Peters.PELLACINI, F., VIDIMČE, K., LEFOHN, A., MOHR, A., LEONE,M., AND WARREN, J. 2005. Lpics: a hybrid hardwareacceleratedrelighting engine for computer cinematography.ACM Transactions on Graphics 24, 3 (Aug.), 464–470.SAKO, Y., AND FUJIMURA, K. 2000. Shape similarity by homotropicdeformation. The Visual Computer 16, 1, 47–61.YEE, Y. L. H. 2000. Spatiotemporal sensistivity and visual attentionfor efficient rendering of dynamic environments. Master’sthesis, Cornell University.4.2 Subdivision on the Geometry Shader5 Array Textures6 Layered Rendering7 Transform Feedback7.1 Physics7.2 Deformation8 ConclusionAcknowledgementsReferencesEXLUNA, INC. 2002. Entropy 3.1 <strong>Technical</strong> Reference, January.FEDKIW, R., STAM, J., AND JENSEN, H. W. 2001. Visual simulationof smoke. In Proceedings of SIGGRAPH 2001, ACMPress / ACM SIGGRAPH, E. Fiume, Ed., Computer GraphicsProceedings, Annual Conference Series, ACM, 15–22.JOBSON, D. J., RAHMAN, Z., AND WOODELL, G. A. 1995.Retinex image processing: Improved fidelity to direct visual observation.In Proceedings of the IS&T Fourth Color ImagingConference: Color Science, Systems, and Applications, vol. 4,124–125.