Image Codec Parallel Optimization Based On Embedded Mobile GPU

Mali-T6xx series of GPU are firstly integrated in the latest ARM CortexA15 series of processors, such series of GPU supports computation framework called OpenCL, OpenCL framework can accelerate general-purpose computing using GPGPU(General-Purpose computation on GPU) technology, the heterogeneous multi-core computation pattern of Mali GPU provides possibility of parallel optimizing directing at multimedia applications such as image encoding/decoding and digital image processing.With the rapid progress of internet and intelligent mobile terminal, the mobile video processing applications which take video encoding/decoding technology as the core are widely used in recent years. The traditional serial computing pattern which relys on CPU efficiency can no longer handle these compute-intensive task with huge massive computation to process, so GPGPU technology has important application value on embedded platforms because embedded platforms need strict limit on hardware cost and power dissipation.Currently the applied method of Mali GPU is lack of researching, the real application example and method summary are especially short of. Therefore, we deeply studied the method of applying GPGPU technology on ARM embedded platform based on Mali-T604 mobile GPU. The research contents include mobile GPU’s hardware structure, the parallel computation pattern of Mali GPU, the programming methods of GPGPU and so on. We achieved the parallel optimization of Jpeg decoding algorithm and Mpeg4 encoding algorithm in open source project LibJpeg and Xvid based on Mali-T604 GPU. We parallelized the computation pattern of the compute-intensive modules such as discrete cosine transform module, pixel interpolating module, motion estimation module, motion compensation module, color space converting module and so on. At last, we tested the optimizing solution actually and analysed the test result.The hardware testing platform is Arndale Board which integrates Mali-T604 GPU and takes Exynos5250 processor as the core. The software testing platform is Ubuntu operating system based on Linux 3.10.37 which is customized for Arndale Board by Linaro. We tested the efficiency of the original serial solution and the optimized solution on encoding/decoding video and image of different resolution and different content, the testing result shows that Mali GPU can promote the efficiency of Jpeg image decoding and Mpeg4 video encoding significantly especially when processing high resolution images, the overall encoding/decoding efficiency was increased 4 times after optimization. The result proves that GPGPU technology base on Mali mobile GPU can provide effective acceleration on multi-media processing application.