T-Buffer: A Dense and Fast Storage for Rendering Order-Independent Transparency

Abstract

To render the transparent effect of a scene fast and in real time, hardware algorithm with additional transparent fragment storage for order-independent transparent fragments are required in graphics processing. As the scene complexity and demand for resolution are constantly increasing, the storage consumption and also its access overhead hurdle system efficiency. This paper proposes the T-buffer storage system design as a solution, which has the following design goals: To lower the demand for transparent fragment storage, we use memory only for pixel locations consisting of transparent fragments. All those fragments with the same pixel location are further grouped in memory for easy management and subsequent use. And to retrieve transparent fragments in computation fast, indexing mechanism is designed in our storage system. We compare the proposed design with two famous works, the R-buffer and the M-buffer with WF hardware oriented algorithm, in terms of storage size and access counts. Experimental results show that T-buffer uses 29% less storage than Rbuffer and 67% less than M-buffer. Nevertheless, T-buffer needs to be accessed 52% less frequent than R-buffer, although 27% more frequent than M-buffer.