Design And Verification Of GPU's 3D Engine Graphics Rendering Pipeline

The development of Graphics Processing Unit(GPU)has experienced three stages:fixed-function Pipelining,Programmable shader and unified Shader Architecture.Because of the very early started research on GPU of western Countries,the Complete Theoretical and Product System had been established decades ago.That is the reason why the western countries take the position of monopoly in several key technologies,such as graphics processing standards,relevant key algorithms and circuit implementation,key processing IP and product implementation.The project,coming from Ministry of China,does the research of relevant techniques on 3D Render Pipeline in order to satisfy the implementation demands on embedded system.This thesis analyzed OpenGL Graphic Pipeline model and the key algorithm of 3D Engine Render Pipeline.According to the analyzing on the characters of implementation,the strategy of 3D Render Pipeline Hardware-Software collaboration had been arranged,and the key-core circuits had been achieved including Vertex Processing,Unified Shader Array,Primitive Assembly,Rasterization,Pixel Processing Unit which could all considered as the key modules of 3D Engine.According to the Structure and functionality on circuit,the verification methods has been proposed.Then based on Simulation Verification platform and FPGA prototype Verification Platform has been established either.The Simulation Verification Platform has been described by different abstract coding languages based on the character of model-level.The highest level has been accomplished by C or C++;the system-c would been used in order to achieve the driver-level.The base-level would been implemented by using behavior modeling language,System Verilog and Verilog,in order to achieve Cycle accurate design and completely achieve the map of all 215 OpenGL Function of 3D Engine.According to hierarchical modelling,the test vectors could be generated automatically which would achieve relevant verification on OpenGL 3D Engine both on function and performance.Those verification methods could highly improve the verification efficiency without any loss of verification accuracy;The FPGA prototype has been accomplished by using two XCVU440 FPGA chip-group coming from XILINX.According to the connection characters of modules,the 3D Engine Render Pipeline logic had been mappedinto above chip-group with the help of EDA Tools.One of those had been used to achieve the function of Unified Shading Array(USA)and Texture Array Unit(TAU).The other one was for the rest functions of 3D Engine.Based on the fully considering of demand of embedded system for 3D Engine Render Pipeline data transfer ability?Vertex processing ability?Geometric processing ability?Pixel processing ability,the specific function and performance verification test had been done.With the benefits from those,the FPGA prototype processing rate reach 30 MHz and the Simulation processing rate reach 600 MHz.Besides that,the pixel filling rat was 4G/s and Tri generating rate was 150M/s.All of those presented that our product satisfied the demands of implementation.