Tile-Based Graphics Processing with Deferred Geometry for Lower Bandwidth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing tile-based graphics processors face inefficiencies in memory bandwidth usage due to the need to store intermediate geometry data generated by geometry processing before binning, which can be optimized by deferring some geometry processing until the rendering stage.
Innovation Solution
The method involves generating geometry packets and deferring certain geometry processing stages until the rendering stage, reducing the need to store intermediate data in memory and optimizing memory bandwidth by performing geometry processing closer to its point of use.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If geometry processing is performed before binning in tile-based graphics processing, then geometry data is available for determining which geometry needs to be processed for respective rendering tiles, but intermediate geometry data must be stored in memory increasing memory bandwidth requirements
Solution Approach 1:
The patent performs a preliminary binning operation before complete geometry processing to generate binning data structures that identify which geometry packets need to be processed for each rendering tile. This preliminary action allows the system to determine geometry-tile relationships without requiring complete geometry processing, thereby reducing the need to store intermediate geometry data in memory and lowering memory bandwidth requirements.
2Manufacturing precision
If all geometry processing is performed before binning, then complete geometry data is available for rendering, but processing time and memory storage requirements increase
Solution Approach 1:
The patent segments the geometry processing pipeline into distinct stages: binning (which operates on geometry packet metadata to generate tile assignments), geometry processing (which processes the actual geometry data), and rendering (which uses the assigned geometry for tile rendering). This segmentation allows binning to proceed independently and in parallel with geometry processing, eliminating the need to wait for complete geometry processing before determining tile assignments, thereby reducing total processing time.
Solution Approach 2:
The binning operation is performed as a preliminary step before complete geometry processing. The binning stage generates data structures that identify which geometry packets need to be processed for each rendering tile, allowing the system to prepare rendering schedules and allocate resources in advance without requiring the actual geometry data to be fully processed yet.
3Productivity
If intermediate geometry data is stored in memory for later processing, then geometry processing can be deferred until rendering stage, but memory bandwidth and storage resources are consumed
Solution Approach 1:
The patent extracts only the essential metadata from geometry packets that is needed for binning operations, such as bounding box information and tile assignment data. This extracted metadata is sufficient to determine which geometry packets need to be processed for each rendering tile, eliminating the need to store and process complete intermediate geometry data in memory. The full geometry data is processed only when needed for actual rendering, reducing memory resource consumption.
Data Source
AI summary
When executing a tile-based graphics processing pipeline, one or more geometry processing stages of a sequence of one or more geometry processing stages of the graphics processing pipeline generate packets storing data for geometry to be processed. A binning stage generates one or more data structures that can be used to determine whether packets should be processed for a rendering tile, and a rendering stage uses the data structures generated by the binning stage to identify packets to be processed for respective rendering tiles. For at least some packets that have been determined as needing to be processed further for a rendering tile, further geometry processing is performed for the packets, and the result of the further geometry processing is used when processing the packets for the rendering tile.