Optimal Matrix Computing Using Vector Division with Sub-word Parallel

Available computing is expected to be effective and efficient in ubiquitous multimedia services. And, matrix multiplication is particular to ubiquitous computing such as image transformation so that it is urgent to exploit potential parallelism as well as data reuse in matrix multiplication. Therefore, based on synchronized data triggered architecture (SDTA), we propose a method of parallel vector division (PVD) as well as sub-word parallel (SWP) supported on SDTA to optimize the matrix computation on specified architecture. Experimental results show that PVD with SWP is suitable for matrix multiplication on SDTA instruction set architecture. Besides, we could extend the proposed model into other data-intensive computing on SDTA.