ARGB Intra-Block Compression for Lower DDR Bandwidth and Power
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The bandwidth limitations of DDR SDRAM in GPUs and video codecs, coupled with high power consumption, are exacerbated by the need to frequently read and write large amounts of image data, which current image block compression algorithms fail to address efficiently due to high algorithm complexity and inadequate utilization of spatial redundancy.
Innovation Solution
A method for lossless ARGB compression using intra-block prediction, combining DPCM and MED predictions to utilize spatial redundancy in two directions, followed by residual grouping and encoding with SVO-VLC, reduces data storage and improves compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If image block compression is employed to reduce data amount for DDR storage, then the access bandwidth requirement is reduced, but the algorithm complexity increases causing higher latency
Solution Approach 1:
The image block is divided into multiple sub-blocks for independent processing. Each sub-block undergoes separate prediction and encoding operations, allowing parallel processing that reduces overall latency while maintaining compression efficiency. The segmentation enables the system to handle complex compression tasks in smaller, more manageable units.
Solution Approach 2:
Prediction values are calculated in advance for all pixels in the image block before the actual encoding process. This preliminary action allows the encoding stage to simply reference pre-computed prediction values and their corresponding residuals, significantly reducing the computational complexity and latency during the main encoding operation.
2Productivity
If complex compression algorithms are used to achieve better compression ratio, then data throughput is improved, but processing speed decreases due to higher latency
Solution Approach 1:
By segmenting the image block into sub-blocks, the system can process multiple sub-blocks in parallel, maintaining high processing speed while applying complex compression algorithms to each segment. This parallel processing approach preserves overall productivity without sacrificing processing speed.
Solution Approach 2:
The patent employs different prediction modes and residual encoding parameters for different sub-blocks based on their local characteristics. This adaptive parameter adjustment optimizes compression ratio for each region while maintaining efficient processing, thereby improving overall data throughput without excessive latency.
3Productivity
If frequent reads and writes to DDR are performed to meet real-time processing demands, then processing capability is maintained, but power consumption increases significantly
Solution Approach 1:
The patent extracts and compresses image data before it is written to DDR, and decompresses it when read. This extraction of compression operations from the main data flow reduces the volume of data transferred to and from DDR, significantly reducing power consumption while maintaining real-time processing capability through efficient compression algorithms.
Data Source
AI summary
A method for lossless ARGB (Alpha, Red, Green, Blue) compression based on an intra-block prediction is provided. The method is executed by a processor, and the method comprises for an input image block under a processing channel, executing the following operations until all input image blocks are encoded: obtaining a predicted value of the input image block under a current processing channel based on the input image block under the current processing channel by the intra-block prediction; determining predicted residuals of the input image block under the current processing channel based on the predicted value and an original pixel value; inputting the predicted residuals into a residual encoder for encoding to obtain a residual stream of the input image block under the current processing channel; and storing residual streams of all input image blocks of all processing channels as a compressed file in a storage file.