Rosearch And Design Of A Tile-based Embedded GPU

Since the21st century, there is a great development of embedded technology and device, which put forward higher requirements on the Graphic Processing Unit (GPU) of an embedded system, especially in the aspects such as drawing speed, drawing effects, hardware consumption. And the development of the Embedded GPU has been vastly promoted.The bandwidth of an embedded system is limited. The processing requirements of low-power, low-cost, high performance and high-quality differs it with the design of desktop GPU. Embedded GPU needs to be optimized in the aspects of architecture design, drawing flow, graphic algorithms, hardware design and the consumption of bandwidth. Tile-based drawing technique is widely used in embedded GPU systems for its high performance to solve these problems. And it is also the main subject of our study.Compared with the traditional drawing pipeline, Tile-based architecture could significantly reduce the bandwidth when the GPU accesses the depth buffer. The main feature of Tile-based drawing is to separate the screen into pixel blocks. Since the blocks are so small that the depth buffer could be placed on-chip. As a result, the frequency that GPU accesses the off-chip memory are obviously decreased, and the bandwidth, power-cost could be diminished.Based on the study of Tile-based drawing methods, the author designs a Tile-based embedded GPU. First, with an intensive study of graphic algorithms, the author chose the suitable algorithms and made an optimization on it, such as the algorithms of the triangle overlap test, Early-Z test and rasterization. Then, the author designed the Tile-based architecture after comparing the real-time rendering with the Tile-based one. The author also finished the design of all hardware modules in the embedded GPU and establishes the verification platform to simulate the design in both module level and system level. The Tile-based embedded GPU proposed in this paper has important meaning of research and can be widely used in the embedded systems to accelerate the3D graphic processing, decrease the consumption of the system bandwidth and power to improve the system performance.