Configurable pixel shader workload reduction technique for mobile GPUs

The pixel shading stage determines substantial shading effects in a programmable graphics pipeline. Owing to the demanding program complexity, the pixel shading stage dominates the rendering performance. In this paper, a configurable pixel shader reduction technique is proposed to enhance the performance for mobile GPUs. Using the developed shading approach, the shading granularity can be configured to construct acceptable, albeit approximated, rendering results to achieve a better trade-off between performance and quality. To estimate the approximation error for determining the rendering quality, a low-order Taylor series expansion is applied in the analysis of the employed shading model. Moreover, a low-overhead hardware architecture is proposed to integrate the configurable shader reduction technique into mobile GPUs with tile-based rasterization architecture. Three levels of shading granularity, i.e., per pixel shading, 2×2 sub-tile shading, and 4×4 sub-tile shading, are supported by configuring rasterizer and ROP unit. Experimental results show that the proposed technique can reach 50% pixel shader reduction ratio for an intermediate shading level, and 70% for a low shading level.