IMEC’s source-code for standard-compliant Scalable Video Decoding (SVC = H.264 extension) now performs 20 times faster and uses only a tenth of the memory footprint compared to the SVC standard reference code. The results were obtained by further optimizing the code through the removal of redundant code and function calls as well as through the improved memory use of IMEC’s earlier (mid-2007) SVC source code. The code was already two times faster than the reference code. The optimized source code is available as starting point for product development by industry via a licensing program and can be delivered as source code.
The optimized decoder has been benchmarked against the reference decoder on a normal end-user PC platform. Compliancy to the SVC standard was also verified. The configurations that were tested included the three scalability types of SVC, being spatial, temporal and quality scalability. The preliminary performance evaluation shows that IMEC’s decoder is up to 20 times faster than the reference software, while consuming only a tenth of the memory.
The first optimization was based on (re)structuring the SVC code into more concise functional blocks. This facilitated extra optimizations, including removing redundant code and function calls. The application of DTSE (Data Transfer and Storage Exploration) transformations enabled intelligent (re)use of the memory footprint. As a result of the increased data locality, the total memory footprint was significantly reduced to one tenth, leading to much better cache behavior and higher performance. This will also contribute to reducing the silicon cost area when using SVC functions for ASIC (application-specific integrated circuit) implementations.
More info: IMEC