Directional Intra Prediction Transposing Reference Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding techniques using directional intra prediction from zone 2 suffer from reduced processing efficiencies due to data being stored in non-contiguous memory locations, leading to decreased overall encoding throughput.
Innovation Solution
The method involves transposing a left reference column of samples into a left reference row, allowing for the computation of predicted tiles based on prediction angles, which are then aggregated to generate a predicted macroblock, thereby reorganizing data into contiguous memory locations for enhanced parallel processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If directional intra prediction from zone 2 is implemented using conventional row-major storage, then the encoding can proceed with standard memory access patterns, but parallel processing efficiency is substantially reduced due to non-contiguous memory locations
Solution Approach 1:
The patent inverts the conventional approach by transposing the left reference column data into a row format before processing. Instead of accessing column data in non-contiguous memory locations during parallel processing, the data is pre-transposed into contiguous row format, allowing efficient parallel access while maintaining the directional intra prediction functionality.
Solution Approach 2:
The patent applies preliminary action by performing the transpose operation on the left reference column data before the parallel processing stage. This pre-processing step reorganizes the data into an optimal format for subsequent parallel operations, eliminating the need for complex memory access patterns during the main encoding process.
2Productivity
If left reference samples are accessed from non-contiguous memory locations during parallel processing, then the directional intra prediction can be performed, but the processing efficiency is substantially degraded
Solution Approach 1:
The transpose operation is performed as a preliminary action before parallel processing begins. By reorganizing the left reference column into a row format in advance, the data is positioned in contiguous memory locations, which dramatically reduces memory access time during the parallel processing stage while enabling efficient directional intra prediction.
3Productivity
If data is stored in row-major order for efficient sequential access, then memory bandwidth utilization is optimized, but column-based operations require non-contiguous access reducing overall efficiency
Solution Approach 1:
The patent inverts the data orientation by transposing column-based reference data into row format. This inversion allows the data to be stored and accessed in contiguous row-major order, which optimizes both sequential access efficiency and parallel processing ease, eliminating the trade-off between these two requirements.
Data Source
AI summary
In various embodiments, a directional intra prediction application encodes video or other media content. The directional intra prediction application transposes a left reference column of samples associated with a first portion of content to generate a left reference row of samples. The directional intra prediction application computes a transposed rightward predicted tile of samples based on a prediction angle and the left reference row of samples. The directional intra prediction application transposes the transposed rightward predicted tile to generate a rightward predicted tile of samples. The directional intra prediction application generates a predicted tile of samples for the first portion of content based on a downward predicted tile of samples and the rightward predicted tile of samples.