Soft IP Cores for parallel video processing on heterogeneous platforms for cloud computing (PHDCOM002)

Parallel programming has recently gained a renewed interest with the raise of cloud computing infrastructures on one side and the availability of high performance computing devices on the other hand such as Graphical Processor Units (GPUs) and Field programmable gate arrays (FPGAs).
Mobile cloud computing paradigms can be leveraged to provide cost-effective support of various applications including video processing (video streaming, transcoding, archiving, retrieval and gaming…etc) over mobile clouds.
In this work, we focus on the development of soft Intellectual Property (IP) cores on heterogeneous platforms combining GPUs and FPGAs to accelerate computationally intensive blocks in video processing such as motion estimation, latent semantic indexing…etc depending on the selected application.
We plan to explore the OpenCL standard together with the new VIVADO tool from Xilinx, which both provide a framework for parallel programming. We also plan to investigate the impact of dynamically reconfigurable co-processor architectures on the performance, energy-efficiency, and resource utilization of video processing systems in mobile cloud environment. We also aim to develop compile-time techniques for managing resources and customizing the heterogonous platform datapath for different stages of program execution. To validate and demonstrate our approach we will use the latest ZinQ and Virtex7 platform to build an application processor prototype that couples a dynamically reconfigurable co-processor with a soft core processor combined with a GPU for application partitioning. System performance, energy-efficiency, resource utilization and video quality metrics will be used for the evaluation of our proposed system.