Optimized SAD calculation algorithm for Cell® processor

Recent multimedia applications demand very high performance to support real-time video encoding. More specifically the mechanism of image block matching is one of the most critical encoder algorithms in terms of computational costs. It is responsible by searching for similar pixel macroblocks when comparing the current macroblock with others in reference images. The SAD (Sum of Absolute Differences) calculation is largelly used as comparation method in different block matching algorithms. This paper purposes a new strategy of SAD method implementation optimized to modern DSP microprocessor architectures like IBM Cell®. Practical results of our proposed method in a H.264/AVC encoder indicate a gain of more than eigth times for each computing unit when compared with original reference code performance.