Graphics Microcontroller Workload Preemption and Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The execution of graphics schedulers on CPUs in computing devices introduces significant latency due to communication delays between graphic engines and the CPU, and traditional methods of tracking unreported or dropped workloads through full memory scanning add additional latency overhead.
Innovation Solution
Implementing a preemption operation in graphics hardware that involves receiving and processing multiple workloads, interrupting lower-priority workloads for high-priority ones, tracking unreported workloads, and re-queuing them after new workloads are loaded, using a graphics microcontroller to manage submit queues and context status reports to minimize latency and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If the graphics scheduler is executed on the CPU to manage multiple graphics workloads, then the graphics engines can be scheduled to execute workloads, but significant latency overhead is introduced due to communication delays between graphic engines and the CPU
Solution Approach 1:
The patent extracts the graphics scheduling function from the CPU and implements it directly in the graphics hardware. The graphics scheduler is integrated into the graphics engine, allowing it to autonomously manage workload execution without requiring continuous CPU intervention or communication over the bus, thereby eliminating the latency overhead associated with CPU-mediated scheduling.
Solution Approach 2:
The graphics hardware performs self-service by internally managing its own workload scheduling. The integrated graphics scheduler within the graphics engine autonomously selects and executes workloads from the submit queue without external CPU control, enabling the system to serve itself and eliminate dependency on CPU timing and communication protocols.
2Reliability
If full scanning of memory registers is used to track unreported or dropped workloads, then all workloads can be accounted for, but additional latency overhead is added
Solution Approach 1:
The patent implements preliminary action by having the graphics hardware automatically track and record dropped workloads during the scheduling process itself. Instead of performing comprehensive memory scanning after the fact, the system proactively monitors workload status transitions and maintains an internal record of dropped workloads, eliminating the need for subsequent full memory scans.
Solution Approach 2:
The patent introduces an intermediary mechanism within the graphics hardware that mediates between workload submission and completion tracking. This internal tracking mechanism acts as a buffer, recording workload status changes and dropped workloads without requiring external CPU scanning operations, thus maintaining reliability while avoiding additional latency.
3Loss of time
If a graphics microcontroller is used to manage submit queues and context status reports, then latency and power consumption are reduced, but device complexity is introduced
Solution Approach 1:
The patent merges the graphics scheduler, submit queue manager, and context status tracker into a single integrated graphics microcontroller unit within the graphics hardware. This consolidation combines multiple functions that would otherwise be separate components into one cohesive unit, reducing the number of communication interfaces and control paths while maintaining functional capability.
Solution Approach 2:
The graphics microcontroller serves multiple functions simultaneously: it manages the submit queue, tracks context status reports, identifies dropped workloads, and coordinates with the CPU. This multi-functional design consolidates what would traditionally require separate dedicated circuits or software modules into a single versatile component, reducing overall system complexity despite the added capability.
Data Source
AI summary
Described herein are technologies related to performing of a preemption operation in graphics hardware. Particularly, the preemption operation includes tracking of unreported or dropped workload independent of workloads submitted to the graphics hardware.