Flexible Uber Fetch Shader Architecture for Fluid Gaming Experience
J. Park, Zhenhong Liu, Sangwon Park, Yongin Kwon, Wilson Fung, Donghee Han · 2025
When running mobile games on RDNA-based GPUs, micro-stutters may occur due to late resource bindings. Vulkan's Graphics Pipeline Library (GPL) separates shading stages for separate compilation, enabling shader compilation during game loading and caching for subsequent use. To enable this, the shader compiler must insert a long instruction sequence, referred to as the Uber Fetch Shader for the Vertex Input Interface, to decouple the vertex buffer format from the vertex attribute fetch code. We propose a novel software/hardware hybrid Vertex Input Interface. This solution improves overall performance and gaming experience during gameplay by reducing compilation time overhead by 16.68% and reducing compilation count by 13.5%.