Parallel Video Codec Pipeline Shutdown After LCU Completion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As video codecs advance to support higher resolutions and frame rates, there is a need for improved power management, particularly to reduce leakage power in video codec parallel processing architectures, as larger Coding Tree Units (CTUs) and Largest Coding Units (LCUs) lead to unbalanced workloads and increased leakage power consumption.
Innovation Solution
Implementing a system that shuts down parallel processing pipelines (pipes) once they complete processing the last LCU row, using interrupt signals to deactivate idle pipes and reduce non-active leakage power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If parallel processing pipelines are used to process video frames with larger LCUs, then processing speed and productivity are improved, but leakage power consumption increases
Solution Approach 1:
The patent applies dynamics by making the processing pipelines dynamically controllable through shutdown/enabled states. The control logic dynamically adjusts pipeline states based on workload completion status, transitioning from static always-on architecture to dynamic state management. This resolves the contradiction by allowing high productivity during active processing while eliminating leakage power when pipelines are idle.
Solution Approach 2:
The patent implements periodic action through interrupt-driven pipeline management. Pipelines are periodically checked for completion status via interrupt signals, and shutdown decisions are made based on periodic workload assessment. This periodic monitoring enables the system to maintain productivity when needed while reducing leakage power during idle periods, resolving the energy-speed tradeoff.
2Reliability
If multiple parallel pipelines are kept active to handle unbalanced workloads, then processing reliability is maintained, but non-active leakage power is consumed
Solution Approach 1:
The patent applies feedback through interrupt signals that provide real-time status information about pipeline completion. The control logic uses this feedback to make informed decisions about pipeline shutdown, ensuring reliability is maintained only when necessary. The feedback mechanism allows the system to distinguish between pipelines that need to remain active (for reliability) and those that can be shut down (to eliminate leakage power).
Solution Approach 2:
The patent implements local quality by applying different states to different pipelines based on their individual workload completion status. Rather than treating all pipelines uniformly, the system applies shutdown logic selectively to completed pipelines while maintaining active state for pipelines still processing work. This localized state management maintains processing reliability where needed while eliminating leakage power from idle pipelines.
Data Source
AI summary
Systems and techniques are provided for processing video data. A process can include obtaining a frame of video data associated with a plurality of Largest Coding Unit (LCU) rows and a plurality of LCU columns, and processing the frame of video data using a plurality of parallel processing pipelines, wherein each pipeline of the plurality of parallel processing pipelines is configured to process a corresponding subset of LCU rows of the plurality of LCU rows. An interrupt signal can be received from a particular pipeline of the plurality of parallel processing pipelines, indicating the particular pipeline has completed processing of a respective last LCU of the corresponding subset of LCU rows. A shut down of the particular pipeline can be caused based on receipt of the interrupt signal from the particular pipeline, wherein the shut down deactivates the particular pipeline.