Parallel Optimization And Realization Of Loeffler Based On Multi-core

With the booming development of computer technology and communication technology, Especially the arriving of big data era, the demand for computer efficiency and compute capability are rising continually. The widespread of multi-core, programming the higher efficient software by it, and replacing the old serial software, tends to be the new trend for software industry. At the same time, as for the traditional serial algorithm, It is significant to enhance the productivity of computer that optimizes and realizes the algorithm based on multi-core calculating platform.This paper, firstly, introduces some key techniques of multi-core, multi-thread, parallel computing which related to fast DCT transform that runs on the multi-core platform, and the mathematic knowledge related to fast DCT; Based on this, describe the compression and decompression briefly, and discuss the entire structure of fast DCT transform. At the same time, through the analyze of loop transform which is the key part of transform, advancing a new decompose method based on Data decompose to optimize Loeffler which is one of the fast DCT transforms, describing the procedure in great detail and analyzing its feasibility.According to the feature of DCT transform, practically the character of2dimension DCT transform, analyze it by using data decompose method. The traditional transform utilizes the double ’for’ loop, in other words, transforming the input step by step through nesting loop. The common method for fast transform utilizes the decompose feature of DCT:row transform fist, then column. This paper aims at Loeffler which is one of the fast DCT transforms, decomposing data deeply, applying the parallelism of Loeffler based on the multi-core. The experiment tells: These three methods showed increasing efficiency at certain condition. Especially the paralleled Loeffler shows the obvious superiority on time at particular condition. As the massive data comes, performance increased dramatically. The experiment shows that the performance enhanced30%at the scale of1600*1200, and efficiency stays stable with the increasing data scale. Above all, the parallelization of Loeffler brings excellent performance according with the experiment expectation.