Optimizing multisample anti-aliasing for deferred renderers

András Máté Fridvalszky

Supervisor(s): Balázs Tóth

Budapest University of Technology and Economics


Abstract: Deferred renderers are popular in computer graphics because they allow using a larger number of light sources, but they have some drawbacks too. One of these is the inability to work together with traditional hardware-based multisample anti-aliasing. Multiple solutions exist to this problem, but their common drawback is the increased memory and bandwidth requirements. We propose a novel approach that eliminates unnecessary memory usage and improves performance while maintaining image quality. Our method is based on a new G-Buffer structure that uses per-pixel linked lists to store the samples. By limiting the number of pre-allocated blocks in the G-Buffer we can also satisfy strict requirements about memory usage and processing time. Similarly to variable rate shading, our method enables to selectively apply anti-aliasing either on preferred parts of the screen or on a per-object basis. We measured the new method using a Vulkan based renderer on scenes with different geometry complexity and characteristics while comparing performance and memory usage to the traditional techniques.
Keywords: Real-time Graphics, Rendering
Full text:
Year: 2020