Exploiting Multi- and Many-core Parallelism for Accelerating Image Compression

With the help of recent development on semiconductor design and process technologies modern processors can provide a great opportunity to increase the performance of processing multimedia data by exploiting task- and data-parallelism in heterogeneous system consisting of multi-core CPUs and many-core GPUs (graphical processing units). This paper presents an optimization of 2D DCT (discrete cosine transform), a computation intensive signal processing algorithm widely used in compression standards, on speed in multi-core CPUs and many-core GPUs. Two optimization techniques using Intel TBB (threading building blocks) and Open CL are discussed in detail. Open CL is a recent open standard proposed to provide universal APIs and programming paradigms for various GPUs and accelerators, it can provide massively parallel processing suitable for data intensive multimedia applications with very low cost. The simulation result that the parallel DCT implementations are performed considerably faster than the serial ones, max 4.8 and 6.9 times speedup as for TBB and Open CL, respectively. Especially, Open CL implementation on GPU shows a linear speedup, a typical characteristic of massively parallel processing, as the increase of 2D data sets.