Area-power efficient lighting unit architecture based on hardware sharing

Abstract

A shared lighting unit architecture suitable for mobile 3D graphics is presented. Sharing of resources is achieved by reusing a unitary vector operation, and diffuse and specular components calculation. Balance between area and power dissipation was compared with a more prevalent parallel unit architecture implementation. Results indicate that parallel implementations of the proposed shared units lead to much better area × power efficiency as more of these units are used for higher processing performance.Introduction: Mobile 3D graphics continuously require more area and power consumption to meet consumer demand for desktop level performance. Since small area and low power is the most important factor in mobile devices, there is a need for well-balanced 3D graphics architecture. In this Letter, we focus on balancing area and power for the lighting calculation. The lighting calculation is one of the most power and area consuming units in the geometric stage of the 3D pipeline [1]. The lighting unit calculates interaction of light sources and modelled objects which are described in terms of vertices and normal vectors. The Open Graphics Library (OpenGL) commonly uses the OpenGL lighting equation (OpenGL LE) to perform the lighting operation by taking into account the ambient, diffuse, and specular reflectance [2]. Typical architectures for lighting calculation are done in parallel, where each reflectance model is developed independent from each other in order to work separately [3,4]. The main drawback of this type of architecture is the large physical area; and since, many of these small lighting processors are needed, the total amount of area consumed is considerably high. Specular and diffuse reflectances require the normalisation of vectors such as the incident lighting ray (L), and the viewer's looking direction (V); it is in this part where most of the area and time are spent, since a series of divisions and square root operations are performed. In this Letter we propose an (area × power) efficient architecture to calculate the diffuse and specular reflectance by sharing the unitary vector operation, and the calculation of the diffuse and specular components. We also present the implementation of the unit vector (UV) block in a parallel architecture to be able to compare both architectures' performance.