Dynamic Load Balancing for Video Decoding

The patent implements dynamic load balancing where the processor automatically adjusts task distribution based on buffer queue levels. When the buffer queue level is high, the processor performs entropy decoding; when low, it performs prediction or deblocking. This dynamic adaptation resolves the contradiction by making the system flexible rather than static, allowing multiple processors to work efficiently without complex manual load balancing.

The system continuously monitors buffer queue levels and uses this feedback to determine which processing task to execute next. This feedback mechanism simplifies multi-processor coordination by providing an automatic control loop that balances workload based on real-time system state, eliminating the need for complex external load balancing algorithms.

The patent segments the video frame into multiple slices, and each slice into multiple macroblock rows. This fine-grained segmentation allows parallel processing at the slice level while using minimal buffer memory. Instead of buffering entire frames for parallel processing, the system processes smaller units independently, resolving the contradiction between parallel processing capability and buffer size requirements.

The patent introduces a new dimension of parallelism by enabling simultaneous processing of multiple slices within a frame.传统方法 only allowed one frame to be processed at a time, but this patent processes multiple slices in parallel across different processor cores, achieving both high productivity and low buffer usage through multi-dimensional parallel processing.

The patent performs preliminary actions by completing entropy decoding and motion compensation before intra-prediction, and by processing macroblock rows in sequence from top to bottom. This preliminary processing establishes the necessary data dependencies upfront, allowing subsequent prediction operations to proceed without complex dependency management. The system prepares data in advance in a way that naturally resolves dependencies.

The patent applies different processing strategies to different parts of the video data. For example, it processes P-frames and B-frames differently, handles intra-prediction and inter-prediction separately, and processes macroblock rows in a specific sequence. This local differentiation resolves data dependency issues by applying context-appropriate processing methods to each region, reducing the need for complex global dependency management.