Fragment Reduction on Mobile GPU with Content Adaptive Sampling

Abstract

Fragment shaders in a graphics pipeline are used to compute the color for each pixel, where lighting, texture loading, and other calculations are involved. The required computing power is proportional to the number of input fragments. In order to improve the power efficiency of mobile GPUs, a content adaptive sampling scheme is proposed to reduce the fragments. The proposed scheme is based on tile-based traversal. For each 4×4 tile, only parts of the fragments are sampled and rendered with the original shader program, and the values of other fragments are interpolated from these rendered fragments if the content checking condition can be passed. With this approach, the sampling patterns are decided adaptively, where more samples are employed for complex regions to avoid quality degradation. Experimental results show that about 30% - 50% fragments can be reduced where high image quality can be still maintained. The proposed scheme can be employed to reduce the power consumption and increase the frame rate for mobile GPUs.