Modern code applied in stencil in edge detection of an image for architecture intel xeon phi knl

Abstract

Modern, high-performance computers are built with a combination of heterogeneous resources, including multi-core and many cores processors, large cache, fast memory, mesh communication between large processes bandwidth, as well as high support for Input/Output capabilities. In order to achieve the best hardware results it is necessary to design highly-performance parallel software with faster modern code that could take full advantage of the vast amount of resources of today’s modern machines. Code modernization encompasses a wide range of activities that aim to improve the performance of highly parallel software. Code modernization is an issue that is being discussed more and more in the field of parallel software development. In this context, the experimentation with Stencil codes through a series of strategies for reorganizing code and algorithms, has shown to increase thread parallelism, vector/SIMD operations and compute intensity in modern architectures. For example, with the use of Stencil codes we have achieved a better performance in image edge detection by a factor of three faster than we could achieved with a standard processor.