Tiled Graphics Workload Scheduling with Unified Command Buffers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graphics processing systems rely on host software to monitor and distribute workloads across multiple tiles, introducing inefficiencies such as command buffer repacketization, increased latency, and power consumption.
Innovation Solution
A work scheduling and submission infrastructure that uses a unified command buffer for workload partitioning, allowing engines within tiles to dynamically acquire and execute work items with minimal latency, eliminating the need for high-level scheduler intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If host software monitors and distributes workloads across multiple tiles, then workload distribution is achieved, but latency and power consumption increase
Solution Approach 1:
The patent extracts the high-level scheduler from the host software and places it directly on the graphics device. This removes the intermediary software layer that causes command buffer repacketization and latency, allowing engines to directly acquire and execute work items from the command buffer without host intervention.
Solution Approach 2:
The graphics device becomes self-sufficient by implementing its own workload scheduling infrastructure. Engines on the device can autonomously monitor the command buffer, acquire work items atomically, and execute them without requiring continuous host software intervention, thereby reducing latency and power consumption.
2Productivity
If host software monitors and distributes workloads across multiple tiles, then workload distribution is achieved, but power consumption increases
Solution Approach 1:
The scheduling function is extracted from the host CPU and transferred to the graphics device itself. This eliminates the need for the host CPU to continuously monitor and redistribute workloads, significantly reducing power consumption while maintaining workload distribution efficiency.
Solution Approach 2:
The graphics device autonomously manages its own workload distribution through on-device scheduling infrastructure. Engines can independently acquire work items from the command buffer without host intervention, eliminating the energy cost of continuous host software monitoring and redistribution.
3Productivity
If command buffer repacketization is performed by host software, then workload partitioning is achieved, but latency and processing overhead increase
Solution Approach 1:
The command buffer is pre-configured with workload partitioning constructs and synchronization commands before submission to the graphics device. This preliminary organization of work items allows engines to directly acquire and execute partitions without requiring runtime repacketization by host software, eliminating processing overhead and latency.
4Productivity
If high-level scheduler intervention is required for workload distribution, then centralized control is maintained, but latency increases
Solution Approach 1:
The scheduling function is segmented from the host software and distributed to individual engines on the graphics device. Each engine can independently acquire work items from the command buffer without requiring centralized scheduler intervention, enabling parallel execution and reducing latency while maintaining controlled workload distribution.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments described herein provide a graphics, media, and compute device having a tiled architecture composed of a number of tiles of smaller graphics devices. The work distribution infrastructure for such device enables the distribution of workloads across multiple tiles of the device. Work items can be submitted to any one or more of the multiple tiles, with workloads able to span multiple tiles. Additionally, upon completion of a work item, graphics, media, and/or compute engines within the device can readily acquire new work items for execution with minimal latency.