VLIW Video Encoding Mask-Based Run Length Loop
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video encoding techniques are inefficient when using Very Long Instruction Word (VLIW) processors for Run-Length Encoding (RLE), Inverse Quantization (IQ), and Variable Length Coding (VLC), as they require serial operations that are not well-suited for VLIW architectures, leading to high cycle consumption, especially at lower bit rates.
Innovation Solution
A mask-based technique that reduces cycle consumption by 30% by combining RLE, IQ, and VLC into a single loop that iterates only over non-zero coefficients, using a mask generated during quantization to optimize operations on VLIW processors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional serial RLE, IQ, and VLC operations are used, then processing correctness is maintained, but processing time and cycle consumption increase significantly
Solution Approach 1:
The patent combines RLE, IQ, and VLC operations into a single integrated loop that processes all three functions simultaneously. The mask generated during quantization is reused across all three operations, eliminating redundant processing and reducing the total number of cycles required compared to conventional sequential implementation.
Solution Approach 2:
The mask is generated during the quantization phase, which is performed before RLE, IQ, and VLC operations. This preliminary generation of the mask allows all subsequent operations to skip zero coefficients efficiently, reducing the processing time for RLE, IQ, and VLC without requiring re-processing of quantized data.
2Productivity
If VLIW architecture is used for parallel processing, then instruction level parallelism is exploited, but serial operations like RLE become less efficient
Solution Approach 1:
The patent merges RLE, IQ, and VLC into a single loop structure that can be efficiently implemented on VLIW processors. By combining these operations and using a pre-generated mask to guide all three processes simultaneously, the patent creates a unified operation that exploits VLIW parallelism while minimizing the inefficiencies associated with serial processing of individual functions.
3Reliability
If RLE processes all coefficients including zeros, then complete encoding is achieved, but redundant operations increase cycle consumption
Solution Approach 1:
The patent extracts and eliminates redundant processing of zero coefficients by using a mask that identifies only non-zero coefficients. The mask is generated during quantization and then used to guide RLE, IQ, and VLC operations to process only necessary data, maintaining encoding completeness while removing wasted cycles on zero coefficient processing.
Solution Approach 2:
The mask generated during quantization serves multiple purposes: it guides RLE to skip zeros, directs IQ to process only non-zero coefficients, and enables VLC to encode only necessary run-length pairs. This self-service approach allows the same data structure to drive multiple operations efficiently without requiring separate processing logic for each function.
Data Source
AI summary
A computer implemented method of video data encoding generates a mask having one bit corresponding each spatial frequency coefficient of a block during quantization. The bit state of the mask depends upon whether the corresponding quantized spatial frequency coefficient is zero or non-zero. The runs of zero quantized spatial frequency coefficients determined by a left most bit detect instruction are determined from the mask and run length encoded. The mask is generated using a look up table to map the scan order of quantization to the zig-zag order of run length encoding. Variable length coding and inverse quantization optionally take place within the run length encoding loop.