Shader Performance Characteristics
GPUs execute fragments in lockstep groups, so a divergent branch costs both paths; lowp/mediump precision qualifiers cut ALU and bandwidth cost on mobile GPUs; and overdraw multiplies fragment cost per pixel, which front-to-back opaque ordering and early depth testing mitigate. Texture fetches inside dynamically computed (non-uniform) control flow also lose automatic derivative-based mip selection.
This Concept is waiting for its first lesson!
GPUs execute fragments in lockstep groups, so a divergent branch costs both paths; lowp/mediump precision qualifiers cut ALU and bandwidth cost on mobile GPUs; and overdraw multiplies fragment cost per pixel, which front-to-back opaque ordering and early depth testing mitigate. Texture fetches inside dynamically computed (non-uniform) control flow also lose automatic derivative-based mip selection.
Are you a teacher? Sign in to start contributing.
Sign In