Somayeh Sardashti - Department of Electical and Computer Eng. University of Tehran
Omid Fatemi - Department of Electical and Computer Eng. University of Tehran

Media processing applications are dominant in many new workloads today. These applications including 3D graphics, image compression, and signal processing, require tens to hundreds of billions of computations per second. To achieve real-time requirements, current media processors use special purpose architectures. Because of the changes in media applications and standards, flexibility is an important factor in new media processors. Hence, programmable media processors are required. Recently, stream processing has been shown to be a suitable candidate for programmable media processors. In this paper, we propose an implementation of motion estimation algorithm onto Imagine, Stanford stream processor. Simulation results show the speed-up of up to 20 compared to sequential processors

Stream Processor, Imagine Processor, SIMD, Motion Estimation