應用於三維繪圖系統之低複雜度切割演算法與高能源效率幾何引擎設計

Abstract

在本篇論文中，我們提出了一個低複雜度的三角形切割渲染演算法與具有高能源效率的幾何引擎架構。所提出的切割演算法與架構能提供近似馮氏渲染的繪圖品質，同時能藉由調整切割層級達到有動態調整繪圖品質與功率消耗的能力。目前可支援三種不同層級的切割模式。 本文所提出的幾何引擎架構運用了數項不同的技術來對功率消耗、效能、面積進行最佳化。例如使用所提出的前向差分切割、邊函數修正、雙空間切割、三角形濾除與頂點快取記憶體等技術可以有效減少幾何轉換與打光運算。針對複雜的運算，我們提出了可重組的資料路徑架構並藉由將資料路徑重組成特定的結構來加速複雜運算的處理。由於重組架構使用相同的硬體進行不同的運算，晶片面積與成本也能有效的減少。與傳統的切割式渲染演算法相比，分別對於一層與兩層的切割渲染運算，我們提出的方法可以減少記憶體存取次數達44.44% 與68.88%，同時幾何轉換乘法運算量也能分別減少50% 與80%。此設計已使用UMC 90 奈米製程實現，從晶片模擬結果顯示 我們所提出幾何引擎可以達到16.978 MVertices/(s*mW)之高能源效率。

In this thesis, a power efficient geometry engine (GE) using a low complexity three-level subdivision algorithm is presented. The proposed subdivision algorithm and architecture is capable of providing low power, scalable and near-Phong shading quality. The forward difference, edge function recovery, dual space subdivision, triangle filtering schemes and post-TnL vertex are employed to alleviate the redundant computation for transforming and lighting of the proposed algorithm and architecture. Due to the three-level subdivision algorithm, one reconfigurable datapth is proposed to reduce the area since the same set of processing elements (PEs) is reused for different operations for the GE. Compared with the conventional subdivision algorithm, the reduction of the number of memory accesses can be attained by 44.44% and 68.88% for level-1 and level-2, respectively. The reduction of the number of multiplications for transforms can be attained by 50% and 80% for level-1 and level-2, respectively. From the implementation results in UMC 90nm, the proposed geometry engine can achieve the power efficiency of 16.978 Mvertices/(s*mW).