高性能GPU模拟器的实现

Implementation of a high performance graphics processor simulator

基于图形处理器(GPU)由固定功能管线与可编程流处理器单元构成的特点,设计了一种半周期精确的模拟器实现方法,并结合一种GPU结构进行了GPU模拟器的实现。GPU模拟器支持OpenGL 2.0 API,在实现了固定功能管线的基础上,支持统一着色器渲染架构。模拟器对GPU结构进行半周期精确模拟。固定功能管线部分实现为完全周期精确模拟,可以进行3D图形算法的正确性验证以及性能评估。对可编程流处理器单元进行指令级功能模拟,在固定功能管线部分的配合下,可以快速轻量地对整个GPU结构进行结构与功能的验证。文章使用模拟器运行了glmark2测试集,并且根据测试结果对GPU结构进行3D图形算法正确性验证与性能评估,结果表明基于本文设计方法的GPU模拟器可以快速轻量地实现3D图形算法正确性验证与性能分析。

Based on the structural characteristics of graphics processing unit(GPU),a half-cycle accurate simulator implementation method is designed.The graphics processor simulator supports the OpenGL 2.0 API and supports a unified shader rendering architecture based on a fixed-function pipeline.The simulator performs a half-cycle accurate simulation of the graphics processor architecture.The fixed-function pipeline part is implemented as a full-cycle accurate simulation.The simulator completes the correctness verification of the graphics algorithm and performance evaluation.Programmable unit is implemented by instruction-level functional simulation.Together with the cooperation of the fixed function pipeline and programmable unit,the simulator can quickly and lightly verify the correctness of the architecture and function of the entire graphics processor architecture.At the end of the article,the simulator runs glmark2,and the performance of the graphics processor architecture is evaluated according to the results,and the feature set is analyzed.It proves that the simulator can achieve the function of correctness verification and architecture performance evaluation quickly and lightly.