雖然圖形處理器的繪圖速度已有長足的進步，在及時繪圖與互動繪圖的應用上還是有不足的地方，對於嵌入式的繪圖處理器而言，更是一大挑戰。在繪圖速度的限制下，程式設計師們無不想方設法地減少繪圖時的運算量，impostor的使用就是一個例子，impostor是一個軟體的機制，物件首先被繪入貼圖中，往後便利用此貼圖來造成繪製原物件的效果，impostor的確大量地減少了計算量。
在這篇論文中，我們看到impostor的顯著效果，也看到了其背後繁複的實作細節，於是我們提出一套硬體的機制，目地是要讓遊戲設計者得到impostor帶來的效能提升，卻不用實作繁複的程式細節，另外，由於我們在硬體上實作，我們可以得到在軟體上無從得知的有用資訊，這使得我們在記憶體的使用上更有效率，也使得此方法可以減少更多的計算量，最後，本篇論文可使整體指令的計算量減少10%以上。

Although the rendering speed of modern GPUs is dramatically improved, it is still not fast enough for some applications such as real time rendering and 3D interactive rendering. Many game developers figure out many methods to reduce the computations of GPUs. One of them is impostor. The impostor method first draw the object into a texture and then apply the texture on a quad or two triangles to generate the illusion of the object. Since the two triangles replace the thousands of triangles, we can reduce a lot of computations.
In this paper, we try to acquire all the benefits of impostor but hide the complex implementation details. Therefore, we propose a hardware mechanism to implement the impostor inside the hardware. With this, game developers do not need to worry about the implementations, because the hardware apply the method automatically. Moreover, since we implement the impostor in the hardware level, we can get some useful data which cannot see in the software. The data help us to apply the impostor technique and use the memory space more efficiently. After all, we can reduce more than 10% number of instructions of the whole GPU system.

1. Introduction 1
2. Previous Work 7
2.1 Billboarding 7
2.2 Impostors 8
2.2.1 What is impostor 8
2.2.2 The motivation of using impostors 9
2.2.3 The software implementations of impostors 9
2.2.3.1 Creating an impostor for an object 10
2.2.3.2 Applying an impostor on two triangles 12
2.3 Existing software approaches to using impostors 15
2.3.1 Dynamically Generated Impostors 15
2.3.2 The use of Impostors in Interactive 3D Graphics System 18
2.3.3 A GPU hardware-based method for automatic occlusion 19
2.4 The issues of textures 20
2.4.1 The limitation of the memory space in the embedded system 20
2.4.2 The 565 texture format 21
2.4.3 Run-length encoding 21
3. Methodology 22
3.1 The state machine and ROM code 24
3.1.1 Phase 1 – the analysis phase 25
3.1.2 Phase 2 – the creation phase 30
3.1.3 Phase 3 – the usage phase 34
3.1.4 Phase 4 – the idle phase 36
4. Experiment Setup 37
4.1 The original number of instructions for GPUs to render the benchmark 37
4.2 The number of instructions for GPUs to render the benchmark 40
5. Results 45
5.1 Performance comparison 45
5.2 GLBenchmark 51
6. Conclusion 54
7. Reference 55

[1] Tomas Akenine-M¨ oller, Eric Haines, Naty Hoffman. Real-Time Rendering(Third Edition)
[2] Frame buffers https://www.open.gl/framebuffers
[3] Billboarding http://www.lighthouse3d.com/opengl/billboarding/
[4] Schaufler, Gernot, “Dynamically Generated Impostors,” GI Workshop on
“Modeling—Virtual Worlds—Distributed Graphics, D.W. Fellner, ed., Infix Verlag,
[5] Kenneth Rohde Christiansen∗, “The use of Imposters in Interactive 3D Graphics Systems” Department of Mathematics and Computing Science Rijksuniversiteit Groningen Blauwborgje 3 NL-9747 AC Groningen
[6] Sheng-Chang Chang, “A GPU hardware-based method for automatic occlusion detection and optimization for objects and subobjects” Department of Computer Science and Engineering National Sun Yat-sen University Master Thesis
[7] GPUs memory latency http://www.sisoftware.net/?d=qa&f=gpu_mem_latency
[8] RGB565 format http://www.theimagingsource.com/en_US/support/documentation/icimagingcontrol-class/PixelformatRGB565.htm
[9] Run-length encoding http://www.fileformat.info/mirror/egff/ch09_03.htm
[10] GLBenchmark http://gfxbench.com/result.jsp