Graphics Processor Fragment Interleaving for Balanced Rendering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional graphics processors exhibit inefficient utilization of rendering resources due to serialised fragment processing workloads, leading to 'bunching' effects and idle periods in functional units, particularly in tile-based rendering systems.
Innovation Solution
Interleaving fragment processing workloads from multiple sources, such as different regions of a render output, to provide a more uniform workload and balance the use of rendering resources by alternating between sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If fragment processing workloads are processed in a serialised manner from single source, then the rendering pipeline is simple to manage, but functional units experience idle periods and resource utilization is inefficient
Solution Approach 1:
The rendering output is divided into multiple regions (e.g., tiles or sub-tiles), and fragment processing workloads are segmented into multiple independent sources corresponding to these regions. Each region can be processed independently, allowing parallel issuance of workloads to functional units, thereby eliminating idle periods and improving resource utilization.
Solution Approach 2:
Multiple fragment processing workloads from different regions are merged into a single interleaved workload stream. The issuance logic combines workloads from multiple sources in an alternating pattern, ensuring continuous utilization of functional units while maintaining manageable complexity through a unified issuance mechanism.
2Productivity
If multiple fragment processing workloads are issued simultaneously from different sources, then resource utilization improves, but workload scheduling complexity increases
Solution Approach 1:
The issuance of fragment processing workloads follows a periodic alternating pattern between multiple sources. For example, workloads are issued in a round-robin fashion or in fixed alternating sequences, providing regular, predictable scheduling that improves resource utilization while avoiding the complexity of dynamic, adaptive scheduling algorithms.
Solution Approach 2:
Workloads from multiple regions are prepared and staged in advance in separate buffers or queues before issuance. This preliminary organization allows the issuance logic to simply alternate between pre-prepared workloads without requiring complex real-time scheduling decisions, thus improving utilization while maintaining simplicity.
3Ease of operation
If fragment workloads are processed sequentially from single source, then processing order is simple to control, but rendering time increases due to idle functional units
Solution Approach 1:
The rendering output is divided into multiple regions (e.g., tiles or sub-tiles), and fragment processing workloads are segmented into multiple independent sources corresponding to these regions. Each region can be processed independently, allowing parallel issuance of workloads to functional units, thereby eliminating idle periods and improving resource utilization.
Solution Approach 2:
The interleaved issuance mechanism ensures that functional units continuously receive fragment processing workloads from multiple sources without idle periods. By alternating between multiple workload sources, the system maintains continuous useful action in the rendering pipeline, reducing total rendering time while keeping control logic simple through systematic alternation.
Data Source
AI summary
There is provided a graphics processor (10) comprising a primitive processing circuit operable to process graphics primitives into respective fragment work items to be rendered by a rendering circuit (22). The primitive processing circuit generates one or more queues (18A, 18B) of fragment work items for rendering that contain fragment work items corresponding to multiple, different sources of fragment work items. The graphics processor (10) is configured to issue fragment work items to the rendering circuit (22) in an interleaved fashion such that rendering of fragment work items from a first source of fragment work items can thereby be interleaved with rendering of fragment work items from a second source of fragment work items.