An Area Efficient Early ${Z}$ -Test Method for 3-D Graphics Rendering Hardware

In this paper, we propose a new early z-test which requires a minimized internal memory while removing redundant z and color reads as well as texture reads. The proposed method determines whether a pixel is screened by a certain mask plane which is containing the history of a pixel's appearance in front of it. If a pixel is screened by the plane, it can be removed. Given an initial position, the method adaptively updates the plane position to maximize the rejected pixels. As a result, on average 39.9% of the memory bandwidth and 21.2% of total power consumption is saved with only a 256 B on-chip memory. The proposed method was implemented into a multimedia system-on-chip with 61 k application-specific integrated circuit (ASIC) gates using a 0.13-mum CMOS process technology.

I. Introduction

Graphics rendering processes for 3-D require lots of memory operations, including texture mapping, stencil tests, tests, color blending, and vertex data fetching. Because of the large memory accesses, the performance of the rendering engine greatly depends on the bandwidth of the memory system. Among the memory operations, most of the operations are required for 3-D rasterization engine (RE) due to the per-pixel processing characteristics. In conventional high-performance rendering processors, texture mapping is performed before the -test [1], [2]. This architecture properly supports the semantics of the standard APIs, such as OpenGL [3], but the major disadvantage is the unnecessary memory operations of texture mapping and -test for hidden pixels, which waste a great deal of the memory bandwidth.