Expansion And Implementation, Based On The Parallelization Of General-purpose Image Library Of Multicore Processors

Nowadays Multi-core architectures(including Many-core near future) will have been the mainstream platform on desktop computing system,the demands on how to fully utilize those parallel resources become an meaningful research topic.Some multi-core programming models on hot and frontier research provides us some feasible method to achieve this target.Ideal method to parallel programming is to fully automatic parallel rewrite those serial programs with parallelism by compilers,but due to the intrinsic complexity of data-stream there is still a long way away from present technical advance.At present the best way to take the programming parallelism is by combining hand-tuning and compiler optimization,it means that programmers should study the parallelism in program and guild compiler to do refactoring work.In this process,the use of libraries is important in parallelism-optimization.Parallel programming has different classification standards:on granularity, fine-grain parallelism and coarse-grain parallelism;from the executing ways,data parallelism and task parallelism,etc.This paper is designed to study coarse-grain data parallelism and task parallelism in multi-core machine.Digital image processing is computing-intensive as well as prevalent and fundamental application.Inside image processing libraries there is rich parallelism existing.we investigate a C++-based generic image processing library(GIL),and manage to add multi-core parallel support on the basis of the mature parallel fundamental library TBB(intel Threading Building Blockings,which provides integral parallel abstract support).Experiments shows,the performance of parallelized GIL have obvious improved on multi-core machines,the original frameworks and interfaces are kept unchanged.