Abstract
In recently, the function of handheld mobile devices become increasingly diverse, many people use handheld devices to play music, video, or picture, clearly the multimedia function is one of the important considerations in purchase handheld devices. However, most embedded systems are still single-core processor architecture, and because of the power considerations, clock rete has also been strictly restricted. Heterogeneous multi-core architecture provides high computation power and also maintains low energy cost. It has become the mainstream embedded systems architecture nowadays. But Android is based on single-core architecture at present stage, the operating system use MPU （main processor unit） to execute all procedures. And the MPU and memory in embedded system device are mostly not effective. When dealing with complex audio and video decoding operations, the multimedia framework can not handle with large data operations. The decoding of video and audio is not smooth. Therefore, how to make improve the utilization of heterogeneous multi-core architecture to enhance multimedia computing has become an important issue on embedded systems. In this paper, we use the platform which is developed by the ITRI. It is a heterogeneous multi-core platform named PAC Duo Platform, composed of an ARM926EJ-S as the MPU, with two PACDSP. PACDSP is a digital signal processor with high performance, low power consumption. It is sufficient to deal with a large number of operations required by multimedia applications work. This paper integrated PACDSP parallel H.264 decoder into Android multimedia framework, including the default multimedia framework- OpenCORE of Android 2.1 versions and the default multimedia framework-Stagefright of Android 2.2. We verify it correctness and performance with Android on PAC Duo Platform. Because the MPU computing load is reduce, experimental results show that H.264 decoding performance can be improve. Our experiment results enable PACDSP decoder-based multimedia framework processes of up to about 53% to 81% faster compared to the original OpenCORE. And 31% to 83% faster compared to original Stagefright. It means through the appropriate distribution of work, decode performance can be enhanced and can maintain the load balancing.