Coarse-Z Filter:A Low-bandwidth Algorithm And Its Hardening For Hidden Surface Removal

Three problems are mainly considered while implementing fixed function pipeline of graphic hardware: First is storage space mainly used for the storage of texture data, depth data, color data and so on, which are needed for object rendering; Second is computation ability including float point operation during geometrical calculation and vertex calculation during raster/ grating; Third is bandwidth ability including bandwidth needed for texture reading cache, depth reading and writing cache and frame reading and writing cache.As the price of storage device quickly goes down and computation ability of GPU quickly comes up, the problems of storage space and computation ability have been effectively eased. However, the bandwidth requirement of graphic hardware is affected by computation ability, commonly, higher computing ability leads to higher bandwidth requirement. Since the development of GPU is much higher than that of graphic hardware interface, nowadays bandwidth ability seems not enough and becomes bottleneck.This paper describes Coarse-Z Filter designed for the optimization of graphic pipeline bandwidth, creates the mathematical model of graphics pipeline bandwidth using Coarse-Z Filter, tests the effectiveness of Coarse-Z Filter with standard industrial testing software according to the model, and analyzes main factors quantitatively, which affects graphics pipeline bandwidth, so as to further estimate the optimistic parameters of Coarse-Z Filter.The core idea of Coarse-Z Filter is to use a low-resolution filtration cache and do Tile level depth testing before pixel level depth testing, so as to reduce the time of Z-Buffer reading. So that graphic pipeline bandwidth can be lowered effectively. Filtration cache stores maximum depth value of all pixels in Tile after scanning and transformation. Coarse-Z Filter has two patterns: single cache pattern and double cache pattern, according to updating strategy of filtration cache.In order to quantitatively analyze contributions of Coarse-Z Filter to optimizationof graphic pipeline bandwidth, this paper creates the bandwidth model of the use of Coarse-Z Filter. According to the bandwidth model, it can be seen that optimization of graphic pipeline bandwidth mainly depends on two factors: efficiency and bandwidth cost of Coarse-Z Filter. Nevertheless, these two factors depends on two parameters Coarse-Z Filter uses: the size of Tile n and bit format of filtration cache k.In order to decide optimistic parameters, this paper does testing on different (n,k) using standard industrial testing software. Testing result shows that in Tile size of 4 X 4 and filtration cache of 16 bits, bandwidth optimization effect of Coarse-Z Filter to graphic pipeline is the most evident. This moment, the use of Coarse-Z Filter can reduce more or less 30% of graphic pipeline bandwidth.The algorithm of Coarse-Z Filter is simple, easy to maintain, easy to be used together with other scene organization algorithms, easy to implement on hardware. And at the same time, it solves the bottleneck of graphics pipeline bandwidth well.